Professional Documents
Culture Documents
Horizons
Pioneering research from the University of Cambridge
Issue 35
Spotlight
Artificial intelligence
Feature
Tree-ring timelines
Feature
Epic poetry
www.cam.ac.uk/research
Issue 35, February 2018 2 Contents
Contents
News Things
4 – 5 Research news 18 – 19 Kettle’s Yard
10 – 11 Taking a moon shot at cystic fibrosis 24 – 25 “Robots can go all the way to Mars...”
12 – 13 Lord of the rings 26 – 27 What’s next for thinking machines?
14 – 15 “Little robots”: behind the scenes at an academy school 28 – 29 From Homer to Hal: 3,000 years of AI narratives
16 – 17 The body in miniature 30 – 31 The malicious use of AI
3 Research Horizons
Welcome
32 – 33 Needles and haystacks: AI in criminology Almost everywhere I turn, I see the transformative potential of
artificial intelligence (AI) being promoted, so it is very timely that
34 – 35 In tech we trust? it is a focus of this issue of Research Horizons.
Some of the researchers featured here are among AI experts
36 – 37 The Cambridge Cluster and AI worldwide who have signed an open letter affirming the benefits
of the technology and urging caution in its development. In
essence, they said: “AI systems must do what we want them
to do.”
This Cambridge Life Enabling enormous promise whilst stewarding progress is
a complex balance. It requires engineers, computer scientists
and mathematicians to build systems that learn from data,
38 – 39 The archaeologist who started her own dig aged seven and that think both like humans and unlike humans; it requires
experts in fields as different as climate science and criminology
to develop innovative uses of these machines that learn; and it
requires researchers to pose new questions about safety, trust,
transparency, security and privacy in an algorithm-rich world.
Cambridge has strengths in machine learning, robotics and
applications of AI technologies. Not only is research aimed at
maximising the impact of AI, it is also aimed at understanding
how we can ensure that the technology benefits humanity. This
has been helped by two new research institutes – the Leverhulme
Centre for the Future of Intelligence and the Centre for the Study
of Existential Risk – as well as being a founding partner in The
Alan Turing Institute.
These developments are indeed timely. In November 2017,
the UK government’s Industrial Strategy set out four Grand
Challenges, one of which was to put the UK at the forefront of
the AI and data revolution. In this issue, we look at some of the
areas in which Cambridge AI researchers are making a significant
impact, as well as consider some of the benefits for academics
and industry of being within the ‘Cambridge Cluster’.
Elsewhere in this varied edition of Research Horizons, we cover
a major boost for cystic fibrosis research, an epic analysis of epic
poetry and Cambridge’s first dedicated tree-ring laboratory.
We hope you enjoy these and other articles in this issue.
Editor
Dr Louise Walsh
Editorial advisors
Dr Mateja Jamnik, Dr Seán Ó hÉigeartaigh,
Dr Beth Singler, Dr Adrian Weller
Design
The District
T +44 (0)1223 765 443
E research.horizons@admin.cam.ac.uk
W cam.ac.uk/research
Copyright ©2018 University of Cambridge and Contributors as identified. The content of Research Horizons, with
the exception of images and illustrations, is made available for non-commercial re-use in another work under the
terms of the Creative Commons Attribution-Non-Commercial-ShareAlike Licence (http://creativecommons.org/
licenses/by-nc-sa/3.0/), subject to acknowledgement of the original author/s, the title of the individual work and
the University of Cambridge. This Licence requires any new work with an adaptation of content to be distributed
and re-licensed under the same licence terms. Research Horizons is produced by the University of Cambridge’s
Office of External Affairs and Communications.
4 News
News
Newton’s papers added
to UNESCO register
Annotated copies of Principia
Mathematica and other papers of
Sir Isaac Newton are now among
materials preserved for the world.
Catching the
memory thief
One of six centres that make up the UK
Dementia Research Institute (DRI) has
opened in Cambridge.
PANI, PAHAR
7 Research Horizons
K
empty Falls is crowded with
tourists who flock to the
nearby Himalayan hill station
of Mussoorie during the summer
months. This stunning beauty spot
lies at the heart of a region beset
by an escalating water crisis.
www.panipahar.com
8 Feature
E
pic poems telling of cultures
colliding, deeply conflicted
identities and a fast-changing
world were written by the Greeks
under Roman rule in the first to
the sixth centuries CE. Now, the
first comprehensive study of these
vast, complex texts is casting
new light on the era that saw the
dawn of Western modernity.
of the written word at all. The vitality of questioning whether anyone truly can
Credit: Wine jar made in Athens around 535 BC. © The Trustees of the British Museum
the spoken word, in the very distinctive command the sea’s depths, a feat that
hexametrical pattern of the poems, was must surely be a journey of the intellect
the single way they had of indicating and imagination. Having acknowledged
authoritative utterance.” the Emperor’s political power, he was,
It is perhaps the most important in effect, implying that the Greeks were
tool available for understanding how perhaps greater masters of knowledge.
the Greeks navigated their loss of The researchers expected to
autonomy under the Romans and find that this tension gave way to a
during the subsequent rise of clearer, moralistic tone, with the rise
Christianity. In recent years, such of Christianity. Instead, they found it
questions have provoked a surge persisted. Nonnus of Panopolis, for
of interest in Greek literature during example, wrote 21 books paraphrasing
that time, but epic poetry itself has the Gospel of St John, but not, it would
largely been overlooked, perhaps seem, from pure devotion, since he also
because it involved large, complex wrote 48 freewheeling stories about
texts around which it is difficult to the Greek god Dionysus. Collectively,
construct a narrative. this vast assemblage evokes parallels
Funded by the Arts and Humanities between the two, not least because
Research Council, Whitmarsh resurrection themes emerge from both.
and his collaborators set out to Nonnus also made much of the son of
systematically analyse the poetry God’s knack for turning water into wine
and its cultural history for the – a subject that similarly links him to
first time. “We would argue Dionysus, god of winemaking.
it’s the greatest gap in ancient Beyond Greek identity itself, the
cultural studies – one of the last poetry hints at shifting ideas about
uncharted territories of Greek knowledge and human nature. Oppian’s
literature,” he adds. poetic guide to fishing, for instance,
The final outputs will include is in fact much more. “I suspect most
books and an edited collection fishermen and fisherwomen know how to
of the poems themselves, but catch fish without reading a Greek epic
the team started simply by poem,” Whitmarsh observes. In fact, the
establishing “what was out there”. poem was as much about deliberately
Astonishingly, they uncovered stretching the language conventionally
evidence of about a thousand texts. used to describe aquaculture, and through
Some remain only as names, others it blurring the boundaries between the
exist in fragments; yet more are vast human and non-human worlds.
epics that survive intact. Together, they Far from just telling stories, then,
show how the Greeks were rethinking these epic poems show how, in an era
their identity, both in the context of the of deeply conflicted identities, Greek
time, and that of their own past and its communities tried to reorganise their
cultural legacy. sense of themselves and their place in
A case in point is Quintus of Smyrna, the world, and give this sense a basis for
author of the Posthomerica – a deceptive future generations. Thanks to Whitmarsh
title since chronologically it fills the gap and his team, they can now be read, as
between Homer’s Iliad and Odyssey, even they were meant to be, on such terms.
communities though it was written later. Quintus’ style “The poetry represents a cultural
therefore was almost uber-Homeric, elaborately statement from the time, but it is also
found themselves crafted to create an almost seamless trying to be timeless,” he adds. “Each
linked closely to connection with the past. Yet there is poem was trying to say something about
their past, while also evidence that, having done so, he also its topic for eternity. The fact that we are
coming to terms with a fast- deliberately disrupted it. “His use of still reading them today, and finding new
metamorphosing future. similes is quite outrageous by Homer’s things to say about them, is a token of
Epic poetry, which standards, for example,” Whitmarsh says. their success.”
many associate with The reason could be Quintus’ painful
Homer’s tales of awareness of a tension between the Professor Tim Whitmarsh
heroic adventure, Homeric past and his own present. Faculty of Classics
seems an odd Conflicted identity is a theme that tjgw100@cam.ac.uk
choice of lens through connects many poems of the period. The
which to examine the poet Oppian, for instance, who wrote Words
transformation. Whitmarsh thinks an epic on fish and fishing, provides us Tom Kirk
its purpose has been misunderstood. with an excellent example of how his
“In the modern West, we often get generation was seeking to reconceive Image
Greek epic wrong by thinking about it as Greek selfhood in the shadow of Rome. Painting on a wine jar of Achilles
a repository for ripping yarns,” he says. The work ostensibly praises the killing Penthesilea, as described
“Actually, it was central to their sense of Emperor as master over land and sea in the epic poem Posthomerica
how the world operated. This wasn’t a – a very Roman formula. Oppian then written by Quintus of Smyrna in
world of scripture; it wasn’t primarily one sabotages his own proclamation by the third century CE
10 Feature
Words
Craig Brierley
A
lmost 30 years on from the scribblings, on the opposite wall books infections that plague people living with
discovery of the genetic defect and files line shelves, and on his desk the condition.
that causes cystic fibrosis, are photos of his family. CF occurs when an individual inherits
treatment options are still limited His desk is somewhat different: it two copies of a single genetic variant,
and growing antibiotic resistance can rise or fall, depending on whether he one from each parent. The disease causes
presents a grave threat. Now, a team wants to work standing or sitting – and a build-up of thick, sticky mucus in the
of researchers from across Cambridge underneath is a treadmill for walking lungs, intestines and organs, and those
hopes to turn fortunes around, thanks and working at the same time. “There affected by the condition are particularly
to a major new centre supported have been times when I’ve been deep in susceptible to lung infections leading to
by the Cystic Fibrosis Trust. thought and almost fallen off it,” he jokes. progressive inflammatory lung damage.
Winn has cystic fibrosis (CF) and Although life expectancy for people with CF
John Winn’s office at Microsoft Research keeping fit is an important part of has almost doubled in recent decades,
looks like that of any typical academic: managing his condition: the stronger his it is still significantly below average.
on one wall is a whiteboard graffitied with lung function, the better equipped he is Winn is a machine learning specialist
impenetrable equations and mathematical to fight the potentially life-threatening and is using his expertise to fight the
11 Research Horizons
condition that affects his everyday resistant and spreading globally. This
life. Together with Professor Andres is one reason why people with CF are A ‘no-strings-attached’ relationship
Floto from Cambridge’s Department of advised not to meet each other.
Medicine, he is turning data from the daily “Clearly the techniques that we
lives of people with CF into potentially develop – and the drug-like molecules Professor Clare Bryant, like Floto, works
life-saving information. that come out of it – will have more on an inflammatory lung disease as
As part of this study, funded by the general applicability to patients with part of the GSK/Cambridge Strategic
Cystic Fibrosis Trust and Papworth other multi-drug resistant infections,” Partnership: in her case, chronic
Hospital, participants have been Floto says. This will be welcome news obstructive pulmonary disease (COPD).
submitting data – everything from heart to England’s Chief Medical Officer,
rate and lung function through to self- Professor Dame Sally Davies, who has COPD is a condition caused by smoking,
reported wellbeing – via an app that also warned of a future where “any one of pollution and severe asthma. Bryant is
monitors their activity levels. Machine us could go into hospital in 20 years for looking in particular at how COPD makes
learning then sifts through the data, minor surgery and die because of an the lungs ‘stickier’ to bacteria, increasing
looking for patterns and – it’s hoped – ordinary infection that can’t be treated the risk of infections.
builds a model that can predict when a by antibiotics.” She holds two grants under the
patient’s health is about to deteriorate The timing of all this is particularly GSK/Cambridge Strategic Partnership,
and advise them to seek medical help. good: Papworth Hospital, whose Adult which aims to develop the next wave of
“The overarching principle is about Cystic Fibrosis Centre has gained a ‘game-changing’ medicines by bringing
giving people control over their own national and international reputation academic and industrial expertise
health data and making it work for them,” for its treatment of patients and its together to tackle often intractable
says Winn. “There’s some informal contribution to research, is due to move disease. Based at Cambridge’s
feedback that just participating in the to the Biomedical Campus later in 2018. Department of Veterinary Medicine,
study and taking these readings has Bryant currently has a three-day-a-week
already improved health outcomes sabbatical at GSK’s headquarters in
for some individuals: for example, Stevenage.
it’s helped with adherence with taking It’s almost 30 years The three-year sabbatical provides
causes CF was
their readings got worse.” GSK resources, but with “no strings
The project is one strand of attached”. The only proviso is that if
research at a major new Cystic Fibrosis
Innovation Hub based on the Cambridge discovered… it’s she works with a GSK reagent, they
have first rights on what she does with
Biomedical Campus and run by Floto.
The Hub is supported through a £5 time to take this shot it. Crucially, she says, it gives her
“the space to think”.
million commitment from the Cystic
Fibrosis Trust and matching funds from
at the moon Bryant is embedded in GSK’s
Respiratory Drug Discovery Unit and
the University of Cambridge. It will attends its lab meeting every week.
strengthen existing collaborations across “I’ve met really smart, clever scientists
the University and with the Wellcome The CF wards will feature state-of-the-art at GSK, with different skills to those
Sanger Institute, as well as build new air flow systems, designed with Floto’s of us in academia,” she says. “I get
collaborative research networks with CF work on the spread of multi-drug resistant to see all aspects of what happens at
centres around the UK. The Trust’s Chief CF pathogens in mind. GSK, everything from how a target is
Executive, David Ramsden, said it will This close proximity between the identified, to how drugs are developed
“provide a step change in CF research patients and the researchers will help Floto to target it, through to taking these
across the country”. test the new treatments he is pioneering. He drugs to clinical trials. I see the
Floto agrees with this sentiment: is particularly excited about the potential for whole spectrum.”
“We have an opportunity to uplift UK new cellular therapies he’s developing with It is, though, a mutually beneficial
CF research in general by providing Professor Ludovic Vallier at the Wellcome- programme, she stresses. Bryant brings
knowhow, training and reagents in a MRC Cambridge Stem Cell Institute. Floto her knowledge of innate immunity and
number of areas including genomics, describes these as their “moon shot”. her experience of multi-disciplinary
bioinformatics, stem cells and clinical These would involve taking cells from a CF collaborations, particularly in imaging.
trials technology.” patient, re-programming them – correcting “It’s effectively like being a consultant,”
A major part of the Hub’s activities the genetic defect along the way – and then she says. “I want them to get as much
will be the development of new drugs re-injecting them into patients. “This could out of me as I do out of them.”
that target chronic inflammation in CF, provide a way to regenerate damaged
in collaboration with the pharmaceutical lungs,” he says.
company GSK as part of the GSK/ Floto knows his plans for the Hub are Professor Clare Bryant
Cambridge Strategic Partnership, as well ambitious, but given that it’s almost 30 Department of Veterinary Medicine
as new antibiotic therapy for the main years since the gene that causes CF was ceb27@cam.ac.uk
causes of lung infection in the condition. discovered and there is still no cure for
Finding new drugs against these the disease, he believes it’s time to take Professor Andres Floto
bacteria is becoming increasingly urgent this shot at the moon. Department of Medicine
– Floto and Professor Julian Parkhill arf27@cam.ac.uk
at the Sanger recently showed that Floto’s collaborators in the CF Innovation
Mycobacterium abscessus, the pathogen Hub include Chris Abell (Chemistry), Dr John Winn
behind one of the most serious infections, Sir Tom Blundell (Biochemistry), Julian Microsoft Research
is becoming increasingly multi-drug Parkhill and Ludovic Vallier. jwinn@microsoft.com
12 Feature
W
hat links a series of volcanic lasted until around AD 660, making this from witnesses who were alive at the
eruptions and severe summer period the coldest experienced during at time – trees. The insight is based on the
cooling with a century of least the last two millennia. It is now known synchronised pattern of ring widths found
pandemics, human migrations, as the Late Antique Little Ice Age, or LALIA. within different tree species at various
political turmoil and the rise and fall Professor Ulf Büntgen, then at the sites across the northern hemisphere.
of civilisations? Tree rings, says Ulf Swiss Federal Research Institute WSL “We believe this exceptionally cold
Büntgen, who leads Cambridge’s and now in Cambridge’s Department of phase from AD 536 to around AD 660 – as
first dedicated tree-ring laboratory Geography, was lead author of the study recorded by very narrow tree rings – should
at the Department of Geography. published in Nature Geoscience in 2016 be considered as a direct or indirect factor
that introduced and described the concept in explaining some of the historical events
AD 536: it’s been called the year that of the LALIA. The team of archaeologists, that occurred both in Europe and Asia
winter never ended. climatologists, geographers and historians during that time,” says Büntgen.
“There was a sign in the sun the like of was the first to provide independent lines As distinctive as a fingerprint, the rings
which had never been seen and reported of absolutely dated and annually resolved formed in trees outside the tropics are
before... The sun became dark and its paleoclimatic evidence for a period of annually precise growth layers. Büntgen
darkness lasted for eighteen months. Each great change that had long perplexed is an expert at assembling, reading and
day it shone for about four hours, and still scientists and historians alike. interpreting these ‘slices of time’ and, since
this light was only a feeble shadow,” wrote “The LALIA coincided with a number his arrival in Cambridge in January 2017,
medieval chronicler Michael the Syrian. of extremely important transformation has set up the University’s first dedicated
A volcanic eruption had thrown a vast processes in human history,” he explains. tree-ring laboratory.
ash cloud into the stratosphere and a “We have the outbreak of the Justinian “You ideally start with a living tree,”
dense fog settled over Europe, the Middle plague across much of the eastern Roman he explains. “This is your anchor point
East and China. It was a year of failed Empire, large-scale migration from inner – you know that the outer layer is this
crops and of famine. Eurasia towards Europe and China, turmoil year’s growth ring, and that the innermost
But worse was to follow. A further two in many parts of central and east Asia, and rings take you back to the tree’s juvenile
volcanic eruptions in 540 and 547 began an the collapse of the eastern Türk Empire.” growth, with the pith ideally referring to
unprecedented cooling across much of the What’s remarkable is that much its birth year. You repeat for many trees,
northern hemisphere. The thermal shock of the evidence for the LALIA comes using statistical analyses to compare
Credit: Hrafn Óskarsson
13 Research Horizons
and match the pattern with other trees “The subfossil wood smells like a fresh precise idea of climatic and environmental
growing at the same time under the same tree, yet this material can be thousands conditions at key periods in history.
environmental conditions, including of years old,” he says. “It’s all about “When you look for links between
climate. Once you’ve gone back as far preservation. If you take wood from a living climate variability and human history
as you can with the oldest living tree you tree and put it in anaerobic conditions you start to build up a multi-dimensional
look for their dead ancestors.” like a lake or in dense clay everything is picture of the past,” he explains. “But the
His team counts rings in the timbers preserved. That’s why we can ultimately subject is overwhelmingly approached
of historical buildings, in subfossil trees compile multi-millennial-long chronologies from within disciplinary silos.”
preserved in bogs and sediments, and in for reconstructing past climate variability.” This is why, since his move to
‘ice-rafted’ driftwood washed up on Arctic Cambridge, Büntgen and colleagues
shores. Back and back they go, comparing from the Department of Geography
and cross-dating, looking for overlaps that
provide new anchor points in the ‘floating
“Once you’ve gone have been forging links with historians,
archaeologists, earth scientists and
chronology’ of patterns. You can see why
Büntgen describes dendrochronology as
back as far as you plant scientists, to make the most of
this remarkable archive.
a big data game.
He is currently involved in a
can with the oldest “Once you embark on these
integrative approaches you can ask
collaborative effort by scientists from
different disciplines and countries to build
living tree you questions like we did for the LALIA –
what was the role of environmental
the world’s longest absolutely dated and look for their dead factors in large-scale human migrations
ancestors”
continuous tree-ring chronology. The team and the rise and fall of ancient
will hopefully soon be able to add another civilisations? How did complex societies
2,000 years, taking the record well into the cope with climate change? That’s when
Late Glacial period near the end of the last it starts to get really exciting.”
major Ice Age around 14,000 years ago. As an environmental scientist, his
This is a huge accomplishment when you main interest is in using continuous tree- Professor Ulf Büntgen
consider that a very cold year might result ring chronologies to reconstruct how the Department of Geography
in a ring that’s only a single cell wide. Earth’s climate system behaved in the ulf.buentgen@geog.cam.ac.uk
His laboratory is full of further past and to understand how ecosystems
collections of wood ready to be analysed, were, and are, responding to temperature Words
including numerous disc samples from and hydroclimatic variation. Louise Walsh
relict larch trees that were discovered in But a timeline as accurate as this
north-eastern Siberia, where hunters look has many other uses, principally in being Image
for mammoth teeth. able to provide a spatially and temporally Drumbabót forest in Iceland
14 Feature
N She found a
ew research lifts the lid on an Prime Minister David Cameron described
influential academy school and academies as “working miracles”.
finds an authoritarian system that
reproduces race and class inequalities.
Primarily state funded but run as not-for-
profit businesses, sometimes with support stress-ridden
‘Structure liberates’: the ethos behind one
from individual philanthropists, academies
such as Dreamfields are independent of hierarchical
of England’s flagship academy schools.
Designed as an engine of social mobility,
local authority control and sit outside the
democratic process of local government. culture focused
this school drills ‘urban children’ for
the grades and behaviour considered a
The gospel according to Dreamfields’
celebrated head is described as a on a conveyer
passport to the world of middle-class
salaries and sensibilities.
“traditional approach”. Kulz says she found
a stress-ridden hierarchical culture focused belt of testing
The headline-grabbing exam results on a conveyer belt of testing under strict –
of this school have led politicians to almost military – conditions, and suffused
champion its approach as a silver bullet with police-style language of ‘investigations’
for entrenched poverty, and ‘structure and ‘repeat offenders’.
liberates’ has become the blueprint for Enforcement comes through what Kulz
recent urban education reform. calls the “verbal cane”. Tongue-lashings
The school’s recipe has now been administered by teachers regularly echoed
replicated many times through academy around the corridors, and were encouraged
trusts that have spread like “modern-day by senior staff. One teacher told Kulz that
missionaries” across the nation, says Dr seeing tall male members of staff screaming
Christy Kulz, a Leverhulme Research Fellow in the faces of 11-year-olds was “very hard
at Cambridge’s Faculty of Education. to digest”.
Shortly after it opened, Kulz was This verbal aggression is heightened
granted permission to conduct fieldwork by the panoptic surveillance built into
in the school, where she had once worked the very architecture of the school. All
as a teaching assistant. Choosing to activity is conducted within the bounds
anonymise her research, she calls the of a U-shaped building with a complete
school Dreamfields. glass frontage. Everyone is on show
Her new book goes behind the scenes at all times, including staff, who felt
of life at Dreamfields, and is the only
detailed ethnographic account of the
everyday practices within this new
breed of academy school.
“Education has long been promoted as
a salve that cures urban deprivation and
balances capitalism’s inequalities,” says
Kulz, who spent 18 months of observation
in Dreamfields, interviewing parents,
teachers and students
“The academy programme taps into
‘mythical qualities’ of social mobility: some
kind of magic formula that provides equal
opportunities for every individual once
they are within the school, regardless of
race, class or social context.” In 2012, then
“Little robots”:
behind the
scenes at an
academy school
15 Research Horizons
constantly monitored and pressured into product quickly and accurately. One student
visibly exerting the discipline favoured by described himself to Kulz as a “little robot”.
management. Most teachers exceeded a 48-hour
Policing was not confined to within the week. The majority of staff were young –
school gates. Kulz goes on a ride-along an average age of 33 – with fewer outside
with what’s known as “chicken-shop commitments, yet many expressed a sense
patrol”. Driving around the streets after of exhaustion. “If you’re not in a lesson we
school, staff members jump out of the car are expected to patrol,” one teacher told
to intervene when children are deemed to Kulz. “Every moment of every day is taken
be congregating or in scruffy uniforms. up with some sort of duty.” Unlike most
Stopping off at one of the local takeaways schools, Dreamfields has no staff room.
is considered a major offence. “Fried chicken Some staff discussed former
represents a ‘poor choice’ that Dreamfields colleagues who had suffered burnout or
must prohibit in order to change urban were asked to resign. During interviews,
culture,” says Kulz. “Simply being caught in Kulz found conspiracy theories were rife
a takeaway after school is punished with a among students because of the number of
two-hour detention the following day.” teachers that “just disappeared”.
Students are also policed through Yet Dreamfields was – and still is –
exacting uniform adherence, with a fêted by politicians and the media for its
‘broken-window theory’ approach that sees undeniably extraordinary exam results: over
deviation as opening the door to chaos. 80% pass rate at GCSE in an area where
this was previously unthinkable. At the time,
the school was vastly oversubscribed, with
over 1,500 applications for just 200 places.
“Most of the students, parents
and teachers were keen to comply to
Dreamfields’ regime, despite its injustices.
The school’s approach was seen as the best
shot at securing grades and succeeding in
an increasingly precarious economy.
“Students, like staff, are trained to be
expendable while the ideals of democracy
and critical thinking we are allegedly meant
to cherish are quashed in the process.”
White middle-class children with long This model of a disciplinarian school
floppy hair, or gathering en masse by built for surveillance and which teaches
Tesco, were ignored. Teachers troubled market-force obedience has marched ever
by this would hint at it in hushed tones. onward since her time in Dreamfields, says
“The approach of many academy Kulz – arriving at new poverty front-lines
schools is one of cultural cloning,” such as rundown seaside towns.
says Kulz. “The Dreamfields creed is Yet, grassroots resistance to this style
that ‘urban children’, a phrase used by of education is increasing. Last year, a
staff to mean working-class and ethnic recently established academy in Great
minority kids assumed to have unhappy Yarmouth that forbade “slouching and
backgrounds, need salvaging – with talking in corridors” had pupils pulled out
middle-class students positioned as the by parents objecting to the “draconian”
unnamed, normative and universal ideal.” rules that were central to the much-imitated
“Black students were consistently Dreamfields playbook.
more heavily policed in the playground, Kulz believes the grades achieved by
resulting in many consciously adopting these schools – far from universally high
The smallest rule infraction can be met ‘whiter’ styles and behaviours – a tactic – come at a price. “We cannot continue
with a spell in isolated detention. that reduced their surveillance.” to ignore the links between the testing
Staff would sometimes go to strange It is not just children who are driven regimes we put pupils through, the harsh
lengths to maintain conformity, she says. hard through incessant monitoring. Staff school cultures they create, and the
Suede shoes were subject to clampdown. at Dreamfields are subject to ‘teacher deteriorating physical and mental health
Parental suggestions of a karaoke stall at tracking’, a rolling system in which student of children and young people in the UK.”
a winter fair were considered far too risky. grades are converted into scores, allowing
“There is no room for unpredictability at management to rank the teachers – an ‘Factories for Learning: Making Race,
Dreamfields,” says Kulz. One student who approach staff compared with salesmen Class and Inequality in the Neoliberal
shaved lines into his eyebrows had to being judged on their weekly turnover. Academy’ (2017) is published by
have them coloured in by a teacher every This pressurised auditing resulted Manchester University Press.
morning. in rote learning to avoid a red flag in the
As fieldwork progressed, however, system. “You put a grade in that satisfies Dr Christy Kulz
Kulz began to notice discrepancies that the system instead of it satisfying the Faculty of Education
tallied uncomfortably with race and social student’s knowledge and needs,” one crk35@cam.ac.uk
background. Black children with fringes, teacher lamented to Kulz, explaining his ‘real
or children who congregated outside job’ was not to teach understanding of his Words
takeaways, were reprimanded immediately. subject, but to get students to produce a set Fred Lewsey
16 Feature
T
THE BODY IN
he past few years have seen of a growing body of work – no pun
an explosion in the number of intended – that uses miniature organ-like
studies using organoids – so- tissues to understand human biology and
called mini-organs – as ways of testing in particular why it goes wrong in cancer
drugs. As the field matures, will we also and dementia. Other research groups
see them being used in personalised in Cambridge are growing mini-brains,
medicine and even in transplants? mini-oesophaguses, mini-bile ducts,
mini-lungs, mini-intestines, mini-wombs,
Dr Laura Broutier reaches into the mini-pancreases… Almost the whole body
incubator and takes out a culture plate in miniature, it seems.
with 24 separate wells, each containing It’s perhaps a misnomer to call
a pale pink liquid. “If you look closely, them mini-organs. They look nothing
you can see the dots there,” she says, like a miniature organ. Rather, they are
manipulating the plates until specks the ‘organoids’, clusters of cells that can
size of a full stop catch the light. grow and proliferate in culture, taking
Broutier is a postdoc in Dr Meritxell on a 3D structure that has the same
Huch’s lab at the Wellcome Trust/Cancer tissue architecture, gene expression
Research UK Gurdon Institute, and these and genetic functions as the part of the
“dots” are miniature liver tumours that have organ being studied.
been regrown from cancer cells taken The technique that Huch uses involves
from patients at nearby Addenbrooke’s taking cells from the liver or, in the case
Hospital. They could make it possible to of her latest work, liver tumours, and
identify cancer drugs personalised for growing these in culture. Her early work
each individual patient. involved growing mini-livers from mouse
Huch’s latest work builds on her stem cells, but she is now working with
previous research on ‘mini-livers’, part human tissue.
MINIATURE
17 Research Horizons
“Organoids have opened up a lot of In the same edition of Development, Cambridge Stem Cell Institute (also a
body in miniature,
made the “most complete human brain are most definitely not unremarkable.
model to date”, he is not as dismissive
it seems
of the field as one might imagine.
The problem, he says, is one of
reproducibility – the same experimental
conditions should yield samples that are
almost identical in terms of size, shape and Such work can currently only be done
composition. This is currently not the case, by transplanting tumour tissue into mice,
he says – organoids can often not be grown growing it over several months and testing
reliably, forcing researchers to ‘cherry pick’ the drugs on the mouse – time-consuming
the best, and even then (and in contrast and technically limiting. Imagine, she
with the organism) each one is different. says, being able to screen hundreds –
“Cells in a Petri dish, like children in even thousands – of drugs at a time on
a playground, will arrange themselves the mini-liver tumours. Clearly this would be
into patterns and some of these will neither practical nor ethical in animals. Dr Laura Broutier
make sense to you. But if we want the “Whether it can be done economically Dr Meritxell Huch
system to be reproducible and useful and practically on an individual patient Wellcome Trust/Cancer
for disease modelling, drug screening basis, time will tell,” she says. “I think, as Research UK Gurdon Institute
or understanding basic mechanisms, with everything, once the technology has mh771@cam.ac.uk
we need to steer them and ensure that if become cheaper, it will be feasible.”
an experiment starts with one hundred It is tempting to speculate that if Professor Alfonso Martinez-Arias
groups of cells, we end up with one scientists can grow organoids in the Department of Genetics
hundred almost identical organoids.” lab, they will soon be able to grow fully ama11@cam.ac.uk
Martinez-Arias’s own work is on functioning organs. But Huch believes
gastruloids – the same concept as we are nowhere near this stage. More Professor Ludovic Vallier
organoids, but used to model very feasible is the idea of using organoids Wellcome-MRC Cambridge Stem
early stages of embryonic development. to replace damaged or diseased tissue Cell Institute
Working closely with physicists and – or more accurately, to help such lv225@cam.ac.uk
engineers, his team has managed to tissue ‘regenerate’. This is one area of
generate gastruloids using mouse research being pursued by Professor Words
cells that are highly reproducible. Ludovic Vallier from the Wellcome-MRC Craig Brierley
18 Things
Things
“The role of art
is to give food
for thought…”
Jim Ede
K
ettle’s Yard – Cambridge
University’s unique modern
art gallery – has re-opened
after an ambitious refurbishment.
Its new research facilities will help
scholars discover why its founder,
Jim Ede, believed “there should be
a Kettle’s Yard in every university.”
www.kettlesyard.co.uk
19
Research Horizons
Credit: Clockwise from top left, Five ships – Mount’s Bay, 1928 (circa) by Alfred Wallis; Jim Ede; the new Kettle’s Yard; letters among Ede’s archive. All images, Kettle’s Yard.
20 Spotlight
LIVING WITH AI
Powerful AI needs to be reliably aligned with human values. Does this mean that AI
will eventually have to police those values? Philosophers Huw Price and Karina Vold
consider the trade-off between safety and autonomy in the era of superintelligence.
21 Research Horizons
T
his has been the decade of AI, the smarts for the job. If there are routes this the future we want, a life in a well-
The uncertain
unicycle that taught
itself and how it’s
helping AI make
good decisions
I 0
0
I
II
0
0
I
I0 0
I
I
0
0I II
0
I
0
II I0
I
0
I
I0 00
I
I
0
0I II
0
0
I
I0 I0
I
0
0
00 0I
I
I
0I I0
0 I 0
0
I
Words
Louise Walsh
23 Research Horizons
C
ambridge researchers are Driverless cars, for instance, may be One of the really exciting frontiers is
pioneering a form of machine trained on a huge dataset of images but being able to model probable outcomes in
learning that starts with only a they might not be able to generalise to the future, as Turner describes. “The role
little prior knowledge and continually foggy conditions. of uncertainty becomes very clear when
learns from the world around it. “Worse than that, the current deep we start to talk about forecasting future
learning systems can sometimes give us problems such as climate change.”
In the centre of the screen is a tiny unicycle. confidently wrong answers, and provide Turner is working with climate
The animation starts, the unicycle lurches limited insight into why they have come to scientists Dr Emily Shuckburgh and Dr
forward and falls. This is trial #1. particular decisions. This is what bothers Scott Hosking at the British Antarctic
It’s now trial #11 and there’s a change me. It’s okay to be wrong but it’s not okay Survey to ask whether machine learning
– an almost imperceptible delay in the fall, to be confidently wrong.” techniques can improve understanding of
perhaps an attempt to right itself before The key is how you deal with climate change risks in the future.
the inevitable crash. “It’s learning from uncertainty – the uncertainty of messy “We need to quantify the future risk
experience,” nods Professor Carl Edward and missing data, and the uncertainty and impacts of extreme weather at a
Rasmussen. of predicting what might happen next. local scale to inform policy responses to
After a minute, the unicycle is gently “Uncertainty is not a good thing – it’s climate change,” explains Shuckburgh.
rocking back and forth as it circles on the something you fight, but you can’t fight it “The traditional computer simulations of
spot. It’s figured out how this extremely by ignoring it,” says Rasmussen. “We are the climate give us a good understanding
unstable system works and has mastered interested in representing the uncertainty.” of the average climate conditions. What
its goal. “The unicycle starts with knowing It turns out that there’s a mathematical we are aiming to do with this work
nothing about what’s going on – it’s only theory that tells you what to do. It was is to combine that knowledge with
been told that its goal is to stay in the first described by 18th-century English observational data from satellites and
centre in an upright fashion. As it starts statistician Thomas Bayes. Ghahramani’s other sources to get a better handle on,
falling forwards and backwards, it starts group was one of the earliest adopters in for example, the risk of low-probability but
to learn,” explains Rasmussen, who AI of Bayesian probability theory, which high-impact weather events.”
leads the Computational and Biological describes how the probability of an event “It’s actually a fascinating machine
Learning Lab in the Department of occurring (such as staying upright in the learning challenge,” says Turner, who is
Engineering. “We had a real unicycle centre) is updated as more evidence (such helping to identify which area of climate
robot but it was actually quite dangerous as the decision the unicycle last took modelling is most amenable to using
– it was strong – and so now we use data before falling over) becomes available. Bayesian probability. “The data are
from the real one to run simulations, and Dr Richard Turner explains how Bayes’ extremely complex, and sometimes missing
we have a mini version.” rule handles continual learning: “the and unlabelled. The uncertainties are rife.”
Rasmussen uses the self-taught system takes its prior knowledge, weights One significant element of uncertainty
unicycle to demonstrate how a machine it by how accurate it thinks that knowledge is the fact that the predictions are based
can start with very little data and learn is, then combines it with new evidence that on our future reduction of emissions, the
dynamically, improving its knowledge is also weighted by its accuracy. extent of which is as yet unknown.
every time it receives new information “This is much more data-efficient than “An interesting part of this for policy
from its environment. The consequences the way a standard neural network works,” makers, aside from the forecasting
of adjusting its motorised momentum and he adds. “New information can cause value, is that you can imagine having a
balance help the unicycle to learn which a neural network to forget everything it machine that continually learns from the
moves were important in helping it to stay learned previously – called catastrophic consequences of mitigation strategies
upright in the centre. forgetting – meaning it needs to look at such as reducing emissions – or the lack
“This is just like a human would learn,” all of its labelled examples all over again, of them – and adjusts its predictions
explains Professor Zoubin Ghahramani, like relearning the rules and glossary of a accordingly,” adds Turner.
who leads the Machine Learning Group language every time you learn a new word. What he is describing is a machine that
in the Department of Engineering. “We “Our system doesn’t need to revisit all – like the unicycle – feeds on uncertainty,
don’t start knowing everything. We learn the data it’s seen before – just like humans learns continuously from the real world,
things incrementally, from only a few don’t remember all past experiences; and assesses and then reassesses all
examples, and we know when we are not instead we learn a summary and we possible outcomes. When it comes to
yet confident in our understanding.” update it as things go on.” climate, however, it’s also a machine of
Ghahramani’s team is pioneering a Ghahramani adds: “The great thing all possible futures.
branch of AI called continual machine about Bayesian machine learning is
learning. He explains that many of the the system makes decisions based on
current forms of machine learning are evidence – it’s sometimes thought of as Professor Zoubin Ghahramani
based on neural networks and deep ‘automating the scientific method’ – and Department of Engineering
learning models that use complex because it’s based on probability, it can zg201@eng.cam.ac.uk
algorithms to find patterns in vast tell us when it’s outside its comfort zone.”
datasets. Common applications include Ghahramani is also Chief Scientist at Professor Carl Edward Rasmussen
translating phrases into different Uber. He sees a future where machines Department of Engineering
languages, recognising people and are continually learning not just individually cer54@eng.cam.ac.uk
objects in images, and detecting unusual but as part of a group. “Whether it’s
spending on credit cards. companies like Uber optimising supply Dr Emily Shuckburgh
“These systems need to be trained on and demand, or autonomous vehicles British Antarctic Survey
millions of labelled examples, which takes alerting each other to what’s ahead on the emsh@bas.ac.uk
time and a lot of computer memory,” he road, or robots working together to lift a
explains. “And they have flaws. When you heavy load – cooperation, and sometimes Dr Richard Turner
test them outside of the data they were competition, in AI will help solve problems Department of Engineering
trained on they tend to perform poorly. across a huge range of industries.” ret26@eng.cam.ac.uk
24 Spotlight: Artificial intelligence
Once an order is received, a robot Sciences Research Council, has just As robots become more common
goes to the shelf where the ordered item completed a three-year project into human– place, in our lives, ethical considerations
is stored. It picks up the shelf and takes it robot interaction, bringing together aspects become more important. In his lab, Iida
to an area where the item is removed and of computer vision, machine learning, public has a robot ‘inventor’, but if the robot
placed in a plastic bin, ready for packing engagement, performance and psychology. invents something of value, who owns
and sending to the customer. It may sound “Robots are not sensitive to emotions the intellectual property? “At the moment,
counterintuitive, but the most difficult part or personality, but personality is the glue the law says that it belongs to the human
of this sequence is taking the item from the in terms of how we behave and interact who programmed the robot, but that’s an
shelf and putting it in the plastic bin. with each other,” she says. “So how do answer to a legislative question,” says Iida.
For Dr Fumiya Iida, this is a typical we improve the way in which robots and “The ethical questions are a little murkier.”
example of what he and other roboticists humans understand one another in a However, philosopher Professor Huw
call a ‘last metre’ problem. “An Amazon social setting?” This is another example of Price, from the Leverhulme Centre for the
order could be anything from a pillow, to a Moravec’s paradox: for most individuals, Future of Intelligence, thinks it will be a long
book, to a hat, to a bicycle,” he says. “For being able to read and respond to the time before we need to think about giving
a human, it’s generally easy to pick up an physical cues of other people, and adapt robots rights.
item without dropping or crushing it – we accordingly, is second nature. For robots, “Think of a dog-lover’s version of the
instinctively know how much force to use. however, it’s a challenge. difference between dogs and cats,” he
But this is really difficult for a robot.” Gunes’ project focused on artificial says. “Dogs feel pleasure and pain, as well
In the 1980s, a group of scientists emotional intelligence: robots that not only as affection, shame and other emotions.
gave this kind of problem another name express emotions, but also read cues and Cats are good at faking these things, but
– Moravec’s paradox – which essentially respond appropriately. Her team developed inside they’re just mindless killers. On this
states that things that are easy for humans computer vision techniques to help robots spectrum, robots are going to be way out
are difficult for robots, and vice versa. recognise different emotional expressions, on the cat end (except for the killing bit,
“Robots can go all the way to Mars, but micro-expressions and human personalities; hopefully) for the foreseeable future. They
they can’t pick up the groceries,” says Iida. and programmed a robot that could come might be good at faking emotions, but
One of the goals of Iida’s lab in across as either introverted or extroverted. they’ll have the same inner life as a teddy
Cambridge’s Department of Engineering is bear or a toaster.
to find effective solutions to various kinds “Eventually we might build robots,
W
e are well on the way to component of human-like computers. His work at the CFI as part of the
a world in which many They will be needed, for instance, so ‘Kinds of Intelligence’ project, led by Dr
aspects of our daily lives that machines can explain their workings Marta Halina, is asking not only what
will depend on AI systems. to humans – an important part of the these tests might look like but also how
transparency of decision-making that we their measurement can be built into the
Within a decade, machines might will require of AI. development of AI. Hernandez-Orallo sees
diagnose patients with the learned With funding from the Engineering a very practical application of such tests:
expertise of not just one doctor but and Physical Sciences Research Council the future job market. “I can imagine a
thousands. They might make judiciary and the Leverhulme Trust, she is building time when universal tests would provide a
recommendations based on vast systems that have begun to reason like measure of what’s needed to accomplish a
datasets of legal decisions and complex humans through diagrams. Her aim now is job, whether it’s by a human or a machine.”
regulations. And they will almost certainly to enable them to move flexibly between Cave is also interested in the impact of
know exactly what’s around the corner different “modalities of reasoning”, just as AI on future jobs, discussing this in a report
in autonomous vehicles. humans have the agility to switch between on the ethics and governance of AI recently
“Machine capabilities are growing,” methods when problem solving. submitted to the House of Lords Select
says Dr Stephen Cave, Executive Director Being able to model one aspect of Committee on AI on behalf of researchers
of the Leverhulme Centre for the Future of human intelligence in computers raises the at Cambridge, Oxford, Imperial College and
Intelligence (CFI). “Machines will perform question of what other aspects would be the University of California at Berkeley.
the tasks that we don’t want to: the useful. And in fact how ‘human-like’ would “AI systems currently remain narrow in
mundane jobs, the dangerous jobs. And we want AI systems to be? This is what their range of abilities by comparison with a
they’ll do the tasks we aren’t capable of – interests Professor José Hernandez-Orallo, human. But the breadth of their capacities
those involving too much data for a human from the Universitat Politècnica de València is increasing rapidly in ways that will pose
to process, or where the machine is simply in Spain and Visiting Fellow at the CFI. new ethical and governance challenges –
faster, better, cheaper.” “We typically put humans as the as well as create new opportunities,” says
Dr Mateja Jamnik, AI expert at the ultimate goal of AI because we have Cave. “Many of these risks and benefits
Department of Computer Science and an anthropocentric view of intelligence will be related to the impact these new
Technology, agrees: “Everything is that places humans at the pinnacle of capacities will have on the economy, and
going in the direction of augmenting a monolith,” says Hernandez-Orallo. the labour market in particular.”
human performance – helping humans, “But human intelligence is just one of Hernandez-Orallo adds: “Much has
cooperating with humans, enabling many kinds. Certain human skills, such been written about the jobs that will be at
humans to concentrate on the areas as reasoning, will be important in future risk in the future. This happens every time
where humans are intrinsically better systems. But perhaps we want to build there is a major shift in the economy. But
such as strategy, creativity and empathy.” systems that ‘fill the gaps that humans just as some machines will do tasks that
Part of the attraction of AI requires cannot reach’, whether it’s AI that thinks humans currently carry out, other machines
that future technologies perform tasks in non-human ways or AI that doesn’t will help humans do what they currently
autonomously, without humans needing think at all. cannot – providing enhanced cognitive
to monitor activities every step of the “I believe that future machines can assistance or replacing lost functions such
way. In other words, machines of the be more powerful than humans not just as memory, hearing or sight.”
future will need to think for themselves. because they are faster but because they Jamnik also sees opportunities in the
But, although computers today can have cognitive functionalities that are age of intelligent machines: “As with any
outperform humans on many tasks, inherently not human.” revolution, there is change. Yes some
including learning from data and making This raises a difficulty, says jobs will become obsolete. But history
decisions, they can still trip up on things Hernandez-Orallo: “How do we measure tells us that there will be jobs appearing.
that are really quite trivial for us. the intelligence of the systems that we These will capitalise on inherently human
Take, for instance, working out the build? Any definition of intelligence qualities. Others will be jobs that we can’t
formula for the area of a parallelogram. needs to be linked to a way of measuring even conceive of – memory augmentation
Humans might use a diagram to it, otherwise it’s like trying to define practitioners, data creators, data bias
visualise how cutting off the corners and electricity without a way of showing it.” correctors, and so on. That’s one reason I
reassembling it as a rectangle simplifies The intelligence tests we use today think this is perhaps the most exciting time
the problem. Machines, however, may “use – such as psychometric tests or animal in the history of humanity.”
calculus or integrate a function. This works, cognition tests – are not suitable for
but it’s like using a sledgehammer to crack measuring intelligence of a new kind, he Dr Stephen Cave
a nut,” says Jamnik, who was recently explains. Perhaps the most famous test Leverhulme Centre for the
appointed Specialist Adviser to the House for AI is that devised by 1950s Cambridge Future of Intelligence (CFI)
of Lords Select Committee on AI. computer scientist Alan Turing. To pass the sjc53@cam.ac.uk
“When I was a child, I was fascinated by Turing Test, a computer must fool a human
the beauty and elegance of mathematical into believing it is human. “Turing never Dr Mateja Jamnik
solutions. I wondered how people came meant it as a test of the sort of AI that is Department of Computer Science
up with such intuitive answers. Today, I becoming possible – apart from anything and Technology (Computer Lab)
work with neuroscientists and experimental else, it’s all or nothing and cannot be used mateja.jamnik@cl.cam.ac.uk
psychologists to investigate this human to rank AI,” says Hernandez-Orallo.
ability to reason and think flexibly, and to In his recently published book The Professor José Hernandez-Orallo
make computers do the same.” Measure of all Minds, he argues for CFI and Universitat Politècnica de
Jamnik believes that AI systems the development of “universal tests of València
that can choose so-called heuristic intelligence” – those that measure the jorallo@dsic.upv.es
approaches – employing practical, often same skill or capability independently
visual, approaches to problem solving – in a of the subject, whether it’s a robot, Words
similar way to humans will be an essential a human or an octopus. Louise Walsh
28 Spotlight: Artificial intelligence
Words
Stuart Roberts
Image
Talos, as imagined in the film
Jason and the Argonauts (1963)
29 Research Horizons
W
e have been writing about Dr Sarah Dillon is Project Lead of the as dominance vs subjugation), the CFI
Credit: Columbia/Kobal/REX/Shutterstock
AI for almost as long as programme – and a devotee of science team is also considering the problems of
stories have been written. fiction and AI storytelling in all its myriad continually perpetuating responses to AI,
Fictions about robots, automatons forms. “All the questions being raised and is thinking of recommendations to
and oracular brass heads have been about AI today have already been explored mitigate against them in a way that creates
with us long before Star Wars’ C-3PO in a very sophisticated fashion, for a very space for more positive – and diverse – AI
and 2001’s killer computer HAL. Now, long time, in science fiction,” says Dillon. narratives to flourish.
researchers want us to consider why “Science fiction literature and film provide To do so, CFI is establishing
the stories we tell ourselves about AI a vast body of thought experiments or partnerships with the wider tech
will have an impact on all our futures. imaginative case studies about what might community as well as engaging with the
happen in the AI future. Such narratives world’s leading AI thinkers from industry,
Nearly 3,000 ago, in the Iliad, Homer ought not to be discarded or derided academia, government and the media.
described Hephaestus, the god of fire, merely because they’re fiction, but rather In December 2017, CFI submitted written
forging women made of gold to serve as thought of as an important dataset. What evidence to the House of Lords Select
his handmaidens – enabling the crippled we want to do is convince everyone how Committee on AI. The AI Narratives
deity to work and move around his forge powerful AI narratives are and highlight programme also includes looking at
underneath Mount Olympus. what effects they can have on our what AI researchers read and how this
In 300 BCE, Apollonius Rhodius everyday lives. People outside of literary influences their research (or not).
imagined Talos, a giant bronze automaton studies have tended not to know how to All this is an attempt by CFI to make
who protected Europa on the Island of deal with this power. sure that future narratives around AI
Crete, in his Greek epic poem Argonautica. “What sort of stories are told – and how aren’t bound by the same prejudices and
And while the term ‘robot’ was only coined they are told – really matters. Fiction has preconceptions as they have been to date.
in the 20th century by Karel Čapek for his influenced science as much as science Says Dillon: “Just consider Google’s
play R.U.R (Rossum’s Universal Robots), has influenced fiction, and will continue to photo app tagging the image of an African
in which artificial servants rise up against do so. One stream of the project is looking American-woman as a gorilla in 2015, or
their masters, we have been imagining directly at how we have talked about new the racist and sexist tweets by Microsoft’s
intelligent machines long before we had technologies in the past – and how we can Chatbot in 2016. If AI continues to learn
the technology capable of creating them. learn from the communication of other our prejudices then the future looks just
Our fascination and appetite for complex technologies when it comes to AI.” as bleak as the past, with the repetition
AI in the pages of our novels, in our Citing the often sensationalist, and consolidation of discrimination and
movie theatres and on our television misinformed or even disingenuous examples inequality.
screens remain undimmed. Two of the of historical narratives around nuclear “Who is telling AI its narratives? Whose
best-received TV shows of recent years energy, genetic engineering and stem stories, and which stories, will inform how
– HBO’s big-budget Westworld and cells, Dillon and her project colleagues AI interacts with the world? Which novels
Channel 4’s Humans – both imagine a Dr Beth Singler and Dr Kanta Dihal suggest are being chosen to ‘teach’ AI morality?
world where AI replicants are on hand that stories around emerging technologies What kind of writers are being enlisted to
to satisfy every human need and desire can significantly influence how they are script AI–human interaction?
– until they reject the ‘life’ of servitude developed, regarded and regulated. “If we can create more diverse literary
they have been programmed to fulfil. Exploring the rich array of themes and cinematic AI narratives, this can
Last autumn, Bladerunner 2049 took associated with AI in history, myth, fiction feed back into the research and into the
cinemagoers into the world originally and public dialogue, the team has been language and data that feeds into actual AI
created by Philip K Dick’s seminal Do unsurprised to find that many pivot around systems. By paying close attention to what
Androids Dream of Electric Sheep? the notion of control: AI as a tool we are stories are doing and how they are doing it,
But how do these old and new, unable to master or a tool that will acquire it doesn’t destroy the power they have – it
polarised and often binary narratives about agency of its own and turn against us. helps us understand and appreciate that
the dawn of the AI age affect, reflect and “The big problem with AI in fiction is power even more.
perhaps even infect our way of thinking dystopia,” says Singler, whose award- “In exploring these AI narratives and
about the benefits and dangers of AI in the winning short documentary film Pain in their concerns, we will be able to bring new
21st century? As the kind of mechanisation the Machine looked at whether robots knowledge derived from literature and film
that existed solely in the minds of should feel pain. “Dystopia can be fun, to current AI debate and hopefully ensure
visionaries such as Mary Shelley, Fritz Lang and people are fascinated by AI, but that the more dystopian futures imagined in
or Arthur C. Clarke looms closer to reality, most of the narratives are written for such narratives do not become our reality.”
we are only just beginning to reflect upon and by young, white men – and that
and understand how such technologies directly influences AI researchers and the
arrive pre-loaded with meaning, sparking research they do. We are not at the stage Dr Kanta Dihal
associations, and media attention, where AI matches human intelligence, Faculty of English and the
disproportionate to their capabilities. but if we do get to a superior form of AI Leverhulme Centre for the
To that end, Cambridge’s Leverhulme or agency, we will find that they too break Future of Intelligence (CFI)
Centre for the Future of Intelligence laws like us. It’s what we do.” ksd38@cam.ac.uk
(CFI) and the Royal Society have come “Isaac Asimov’s legendary Four Laws
together to form the AI Narratives research of Robotics, for example, have become so Dr Sarah Dillon
programme. It’s the first large-scale ubiquitous that they were referenced in a Faculty of English and CFI
project of its kind to look at how AI has, 100-page report by the US Navy, which is sjd27@cam.ac.uk
and is, being portrayed in popular culture slightly terrifying,” says Dihal. “The Laws
– and what impact this has not only on are a storytelling device. If Asimov’s Laws Dr Beth Singler
readers and movie-goers, but also on worked perfectly there would be no story!” Faraday Institute for Science
AI researchers, military and government As well as identifying recurrent and Religion and CFI
bodies, and the wider public. dichotomies in popular AI narratives (such bvw20@cam.ac.uk
30 Spotlight: Artificial intelligence
T
he future used to belong to The scenario above never happened. robots and drones that may be designed
science fiction writers. But the Or at least, it hasn’t happened yet. or repurposed for attacks – as well as an
technologies once imagined But it is one of several possible real-life unprecedented rise in the use of ‘bots’
by Philip K. Dick and Ray Bradbury scenarios envisaged by some of the to manipulate everything from elections
now belong to the realm of science world’s leading experts on the impacts of and the news agenda to social media. It
fact. What visions of the future AI – who joined forces to author and sign issues a clarion call for governments and
might the world’s leading AI experts a ground-breaking report that sounds the corporations worldwide to address the
predict if you put them in a room alarm about the potential future misuse of clear and present danger inherent in the
together? Cambridge’s Centre for AI by rogue states, terrorists and malicious myriad applications of AI.
the Study of Existential Risk (CSER) groups or individuals. In addition, the report – The
and Oxford’s Future of Humanity The report forecasts dramatic growth Malicious Use of Artificial Intelligence:
Institute decided to find out… during the next decade in the use of Forecasting, Prevention, and Mitigation
31
31 Research Horizons
– also identifies potential solutions Likewise, the proliferation of cyber- hyped promises failed to match the
and interventions to allay some of the physical systems will allow attackers to reality of how difficult it has been to make
potentially catastrophic risks discussed. deploy or repurpose such systems for progress on these technologies,” explains
Experts in the fields of machine learning, harmful ends (such as turning commercial Avin. “With all the rapid progress in
AI safety, drones, cybersecurity, lethal drones into face-targeting missiles or recent years, brought about in part by
autonomous weapons systems and holding critical infrastructure to ransom). much more capable processors, there
counterterrorism, from organisations The rise of autonomous weapons systems has yet to be a clear point of maturation,
such as Google, OpenAI, DeepMind and in the battlefield also risks the loss of of acknowledging that this technology is
Microsoft, as well as leading thinkers from meaningful human control and increases going to change everyone’s lives this time
Cambridge, Yale, Oxford and Princeton the prospects of targeted autonomous around, and we need to start planning for
Universities (among others), came together attacks. the potential risks and benefits.”
in Oxford to address the critical challenges Meanwhile, in the political sphere, Avin and Ó hÉigeartaigh suggest that
around AI in the 21st century. detailed analytics and the automation CSER is uniquely placed to contribute
Together, the participants highlighted of message creation present powerful to discussions around the study and
important changes to the strategic security tools for manipulating public opinion mitigation of risks associated with
landscape, which could include: more on previously unimaginable scales. emerging technologies and human
attacks, due to the scalable automation of “The aggregation of information activity. For the purpose of this report,
attacks; harder to defend against attacks, by states and corporations, and the this meant being able to convene experts
due to the dynamic nature of AI; and increasing ability to analyse and act in machine learning, cybersecurity
more attackers, as skill and computing on this information at scale using AI and the broader legal, socio-political
resources become increasingly available. could enable new levels of surveillance implications. The result is a report
“The consequences of such and invasions of privacy, and threaten that lays out how and why AI will alter
developments are difficult to predict in to radically shift the power between the landscape of risk for citizens,
detail, and not all participants agreed individuals, corporations and states,” organisations and states.
on all conclusions,” says Dr Shahar Avin adds Ó hÉigeartaigh. “It is often the case that AI systems
of CSER who co-chaired the workshop don’t merely reach human levels of
with Miles Brundage from Oxford. performance but significantly surpass
“However, there was broad consensus on
predictions around attacks that are novel
“AI is a game them,” says the report. “It is troubling,
but necessary, to consider the
either in the form of attacks on AI systems
or because they are carried out by AI
changer and this implications of superhuman hacking,
surveillance, persuasion, and physical
systems; more targeted attacks, through
automated identification of victims;
report has imagined target identification, as well as AI
capabilities that are subhuman but
and unattributable attacks through AI
intermediaries.”
what the world could nevertheless much more scalable than
human labour.”
“AI is a game changer and this report look like in the next Adds Ó hÉigeartaigh: “Whether it’s
has imagined what the world could criminals training machines to hack or
look like in the next five to ten years,” five to ten years” ‘phish’ at human levels of performance
adds Dr Seán Ó hÉigeartaigh, Executive or privacy-eliminating surveillance and
Director of CSER and one of the report profiling – the full range of impacts on
signatories. “We live in a world that could To mitigate such risk, the authors security is vast.”
become fraught with day-to-day hazards explore several interventions to reduce
from the misuse of AI and we need to take threats associated with the malicious ‘The Malicious Use of Artificial
ownership of the problems – because the use of AI. They include recommendations Intelligence: Forecasting, Prevention, and
risks are real. There are choices that we for more engaged policy making and Mitigation’ is the result of a workshop co-
need to make now. Our report is a call more responsible development of the organised by CSER and the University of
to arms for governments, institutions technology, an opportunity to learn from Oxford’s Future of Humanity Institute.
and individuals across the globe.” the best practices of other risky fields,
He adds: “For many decades, hype and a call for a “broader conversation”.
was outstripping fact in terms of AI and The report also highlights key areas
machine learning. Now, the situation for further research, including at the
is being reversed and we have to rethink intersection of AI and cybersecurity, on
all the ways we currently do things. openness and information sharing of risky
This report looks at the practices that just capabilities, on the promotion of a culture
don’t work anymore – and suggests broad of responsibility, and on seeking both
approaches that might help: for example, institutional and technological solutions
how to design software and hardware to to tip the balance in favour of those
make it less hackable – and what type of defending against attacks. Dr Shahar Avin
laws and international regulations While the design and use of Centre for the Study of
might work in tandem with this.” dangerous AI systems by malicious Existential Risk (CSER)
The report also identifies three actors has been highlighted in high- sa478@cam.ac.uk
security domains (digital, physical and profile settings (such as the US Congress
political) as particularly relevant to the and White House, separately), the Dr Seán Ó hÉigeartaigh
malicious use of AI. It suggests that AI intersection of AI and malicious use on a CSER
will disrupt the trade-off between scale massive scale has not yet been analysed so348@cam.ac.uk
and efficiency, and allow large-scale, comprehensively – until now.
highly efficient and targeted attacks “The field of AI has gone through Words
on digital systems. several so-called ‘winters’, when over- Stuart Roberts
32 Spotlight: Artificial intelligence
NEEDLES &
HAYSTACKS
Police at the “front line” of difficult
risk-based judgements are trialling an AI
system trained to give guidance using the
outcomes of five years of criminal histories.
Words
Louise Walsh
33 Research Horizons
I
“ t’s 3am on Saturday morning. The new serious offence such as murder, “Not all errors are equal,” says Sheena
IN
TECH
WE
TRUST?
Fairness, trust and transparency are qualities we usually associate
with organisations or individuals. Today, these attributes might also
apply to algorithms. As machine learning systems become more complex
and pervasive, Cambridge researchers believe it’s time for new thinking
about new technology.
35 Research Horizons
D
r Jat Singh is familiar with machine learning is changing the way Googling her name, she was shocked to
breaking new ground and we live and work. be presented with ads suggesting that she
working across disciplines. Even “Not long ago, many markets were had been arrested. After much research,
so, he and colleagues were pleasantly traded on exchanges by people in pits she discovered that “black-sounding”
surprised by how much enthusiasm screaming and yelling,” Weller recalls. names were 25% more likely to result in
has greeted their new Strategic “Today, most market making and order the delivery of this kind of advertising.
Research Initiative on Trustworthy matching is handled by computers. Like Sweeney, Weller is both disturbed
Technologies, which brings together Automated algorithms can typically provide and intrigued by examples of machine-
science, technology and humanities tighter, more responsive markets – and learned discrimination. “It’s a worry,” he
researchers from across the University. liquid markets are good for society.” acknowledges. “And people sometimes
But cutting humans out of the loop stop there – they assume it’s a case of
In fact, Singh, a researcher in Cambridge’s can have unintended consequences, as garbage in, garbage out, end of story.
Department of Computer Science and the flash crash of 2010 shows. During In fact, it’s just the beginning, because
Technology, has been collaborating 36 minutes on 6 May, nearly one trillion we’re developing techniques that can
with lawyers for several years: “A legal dollars were wiped off US stock markets automatically detect and remove some
perspective is paramount when you’re as an unusually large sell order produced forms of bias.”
researching the technical dimensions to an emergent coordinated response from Transparency, reliability and
compliance, accountability and trust in automated algorithms. “The flash crash trustworthiness are at the core of Weller’s
emerging ICT; although the Computer Lab was an important example illustrating work at the Leverhulme Centre for the
is not the usual home for lawyers, we have that over time, as we have more AI agents Future of Intelligence and The Alan Turing
two joining soon.” operating in the real world, they may Institute. His project grapples with how
Governance and public trust present interact in ways that are hard to predict,” to make machine-learning decisions
some of the greatest challenges in he says. interpretable, develop new ways to ensure
technology today. The European General Algorithms are also beginning to be that AI systems perform well in real-world
Data Protection Regulation (GDPR), involved in critical decisions about our lives settings, and examine whether empathy is
which comes into force this year, has and liberty. In medicine, machine learning is possible – or desirable – in AI.
brought forward debates such as helping diagnose diseases such as cancer Machine learning systems are here
whether individuals have a ‘right to an and diabetic retinopathy; in US courts, to stay. Whether they are a force for
explanation’ regarding decisions made by algorithms are used to inform decisions good rather than a source of division
machines, and introduces stiff penalties about bail, sentencing and parole; and on and discrimination depends partly on
for breaching data protection rules. “With social media and the web, our personal researchers such as Singh and Weller.
penalties including fines of up to 4% of data and browsing history shape the news The stakes are high, but so are the
global turnover or €20 million, people stories and advertisements we see. opportunities. Universities have a vital
are realising that they need to take data How much we trust the ‘black box’ role to play, both as critic and conscience
protection much more seriously,” he says. of machine learning systems, both of society. Academics can help society
Singh is particularly interested in how as individuals and society, is clearly imagine what lies ahead and decide what
data-driven systems and algorithms – important. “There are settings, such as we want from machine learning – and what
including machine learning – will soon criminal justice, where we need to be it would be wise to guard against.
underpin and automate everything from able to ask why a system arrived at its Weller believes the future of work is a
transport networks to council services. conclusion – to check that appropriate huge issue: “Many jobs will be substantially
As we work, shop and travel, computers process was followed, and to enable altered if not replaced by machines in
and mobile phones already collect, meaningful challenge,” says Weller. coming decades. We need to think about
transmit and process much data about us; “Equally, to have effective real-world how to deal with these big changes.”
as the ‘Internet of Things’ continues deployment of algorithmic systems, And academics must keep talking
to instrument the physical world, machines people will have to trust them.” as well as thinking. “We’re grappling
will increasingly mediate and influence But even if we can lift the lid on these with pressing and important issues,” he
our lives. black boxes, how do we interpret what’s concludes. “As technical experts we need
It’s a future that raises profound issues going on inside? “There are many kinds to engage with society and talk about what
of privacy, security, safety and ultimately of transparency,” he explains. “A user we’re doing so that policy makers can try
trust, says Singh, whose research is contesting a decision needs a different to work towards policy that’s technically
funded by an Engineering and Physical kind of transparency to a developer who and legally sensible.”
Sciences Research Council Fellowship: wants to debug a system. And a third form
“We work on mechanisms for better of transparency might be needed to ensure
transparency, control and agency in a system is accountable if something goes
systems, so that, for instance, if I give wrong, for example an accident involving Dr Jat Singh
data to someone or something, there are a driverless car.” Department of Computer Science
means for ensuring they’re doing the right If we can make them trustworthy and and Technology (Computer Lab)
things with it. We are also active in policy transparent, how can we ensure that js573@cam.ac.uk
discussions to help better align the worlds algorithms do not discriminate unfairly
of technology and law.” against particular groups? While it might be Dr Adrian Weller
What it means to trust machine learning useful for Google to advertise products it Department of Engineering,
systems also concerns Dr Adrian Weller. ‘thinks’ we are most likely to buy, it is more the Leverhulme Centre for the
Before becoming a senior research fellow disquieting to discover the assumptions it Future of Intelligence and
in the Department of Engineering and a makes based on our name or postcode. The Alan Turing Insititute
Turing Fellow at The Alan Turing Institute, When Latanya Sweeney, Professor of adrian.weller@eng.cam.ac.uk
he spent many years working in trading Government and Technology in Residence
for leading investment banks and hedge at Harvard University, tried to track Words
funds, and has seen first-hand how down one of her academic papers by Becky Allen
36 Spotlight: Artificial intelligence
37 Research Horizons
O
n any given day, some of the “Principled AI is almost an old-fashioned where the best commercial applications
world’s brightest minds in the way of thinking about the world,” says are for their platform.
areas of AI and machine learning Chatrath. “Humans are capable of making AI and machine learning companies
can be found riding the train between good decisions quickly, and probabilistic like PROWLER.io are clearly tapping into
Cambridge and London King’s Cross. models like ours are able to replicate that, what could be a massive growth area
but with millions of data points. Data isn’t for the UK economy: PwC estimates that
Five of the biggest tech companies in the king: the model is king. And that’s what AI could add £232 billion to the economy
world – Google, Facebook, Apple, Amazon principled AI means.” by 2030; and the government’s Industrial
and Microsoft – all have offices at one Could PROWLER.io be the next Strategy describes investments aimed
or both ends of the train line. Apart from big success story from the so-called at making the UK a global centre for AI
the tech giants, however, both cities (and ‘Cambridge cluster’ of knowledge- and data-driven innovation. But given
Oxford, the third corner of the UK’s so- intensive firms? In just under two years, the big salaries that can come with a
called golden triangle) also support thriving the company has grown to more than career in big tech, how can universities
ecosystems of start-ups. Over the past 60 employees, has filed multiple patents prevent a ‘brain drain’ in their computer
decade, start-ups based on AI and machine and published papers. Many of the science, engineering and mathematics
learning, in Cambridge and elsewhere, have people working at the company have departments?
seen explosive growth. deep links with the University and its The University has a long tradition of
Of course, it’s not unexpected that research base, and many have worked entrepreneurial researchers who have
a cluster of high-tech companies would for other Cambridge start-ups. Like any built and sold multiple companies while
sprout up next to one of the world’s new company, what PROWLER.io needs maintaining their academic careers,
leading universities. But what is it that to grow is talent, whether it’s coming running labs and teaching students.
makes Cambridge, a small city on the from Cambridge or from farther afield. “People from academia are joining us
edge of the Fens, such a good place to “There’s so much talent here already, and feeding back into academia – in
start a business? but it’s also relatively easy to convince Cambridge, there’s this culture of ideas
“In my experience, Silicon Valley is 10% people to move to Cambridge,” says going back and forth,” says Chatrath.
tech and 90% hype, but Cambridge is just Rasmussen. “Even with the uncertainty “Of course some people will choose to
the opposite,” says Vishal Chatrath, CEO pursue a career in industry, but Cambridge
of PROWLER.io, a Cambridge-based AI has this great tradition of academics
company. “As an entrepreneur, I want to
bring world-changing technology to market. “In my experience, choosing to pursue both paths – perhaps
one will take precedence over the other for
The way you do that is to make something
that’s never existed before and create the Silicon Valley is 10% a time, but it is possible here to be both an
academic and an entrepreneur.”
science behind it. Cambridge, with its rich
history of mathematicians, has the kind of tech and 90% hype, “I don’t know of any other university
in the world that lets you do this in terms
scientific ambition to do that.”
“The ecosystem in Cambridge is but Cambridge is of IP. It’s a pretty unique set-up that I can
start a business, raise venture capital,
really healthy,” says Professor Carl
Edward Rasmussen from Cambridge’s just the opposite” and still retain a research position and
do open-ended research. I feel very
Department of Engineering, and Chair of lucky,” says Dr Alex Kendall, who recently
PROWLER.io. “The company has been completed his PhD in Professor Roberto
expanding at an incredible rate, and that comes along with working for a Cipolla’s group in the Department of
I think this is something that can only start-up, there’s so much going on Engineering, as well as founding Wayve,
happen in Cambridge. here that even if a start-up isn’t a Cambridge-based machine learning
PROWLER.io is developing what ultimately successful, there are always company. “A lot of other universities
it calls the world’s first ‘principled’ AI new opportunities for talented people wouldn’t allow this, but here you can –
decision-making platform, which could because the ecosystem is so rich.” and it’s resulted in some pretty amazing
be used in a variety of sectors, including “Entrepreneurs in Cambridge really companies.”
autonomous driving, logistics, gaming support one another – people often call “I didn’t get into this field because I
and finance. Most AI decision-making each other up and bounce ideas around,” thought it would be useful or that I’d start
platforms tend to view the world like says Carol Cheung, an Investment lots of companies – I got into it because
an old-fashioned flowchart, in which Associate at Cambridge Innovation I thought it was really interesting,” says
the world is static. But in the real world, Capital (CIC). “You don’t often see that Professor Zoubin Ghahramani, one of
every time a decision is made, there are degree of collaboration in other places.” Cambridge’s high-profile entrepreneurial
certain parameters to take into account. CIC is a builder of high-growth academics, who splits his time between
“If you could take every decision- technology companies in the Cambridge the Department of Engineering and his
making point and treat it as an autonomous Cluster, and has been an important Chief Scientist role at Uber. “There were
AI agent, you could understand the addition to the Cambridge ecosystem. It so many false starts in AI when people
incentives under which the decision is provides long-term support to companies thought this is going to be very useful
made,” says Chatrath. “Every time these that helps to bridge the critical middle and it wasn’t. Five years ago, AI was like
agents make a decision, it changes the stage of commercial development – the any other academic field, but now it’s
environment, and the agents have an ‘valley of death’ between when a company changing so fast – and we’ve got such
awareness of all the other agents. All first receives funding and when it begins a tremendous concentration of the right
these things work together to make the to generate steady revenue – and is kind of talent here in Cambridge to take
best decision.” a preferred investor for the University advantage of it.”
For example, autonomous cars of Cambridge. One of CIC’s recent
running PROWLER.io’s platform would investments was to lead a £10 million
communicate with one another to alleviate funding round for PROWLER.io, and it Words
traffic jams by re-routing automatically. will work with the company to understand Sarah Collins
38 This Cambridge Life
Robotics
making appearances and raising money
for good causes – and I’ve learned new
skills, everything from how to trim plastic
armour to how to wire electrics and how
to airbrush alien headdresses!
cambridge.org/core-robotics
Dr Jennifer Bates
Department of Archaeology
jb599@cam.ac.uk
Interview
Alex Buxton
T +44 (0)1223 765 443 Contact