16 MAY2020|COMPUTERSHOPPER|ISSUE
It’snotforus
toadaptto
thiscompany,it’s
forthiscompanyto
adapttous”
EuropeanCommissionerThierry
BretononFacebook’sdesireto
dictatehowitshouldberegulated
TheColonel
waspassionate
about his fried
chicken and this
newSkill makes
re-ordering your
favourite KFC menu
items easier than
ever but in afun and
memorable way”
JasonCassidy,marketingdirector
forKFCCanada, onthe fast-food
joint’s adoption ofAmazon Alexa
Creating
unique and
previously
impossible
opportunities for
private citizens to
experience space
is whySpace
Adventures exists”
SpaceAdventureschairmanEric
Andersononapartnership with
ElonMusk’s SpaceX
Forthose who
fail to embrace
change,thereisa
real risk of being
left behind”
Baroness Martha Lane-Fox
highlights the danger of adigital
skills shortage in the UK
Adigital form
requiring
manual input might
be moreaccurate,
quickerand cheaper”
Agovernment guideto AI use in
the public sector misses the point
ARTIFICIAL INTELLIGENCE(AI)technology could
be used to rewriteoutdated Wikipedia articles,
reducing the workload forhuman editors, thanks
to asystemdeveloped by Massachusetts Institute
of Te chnology (MIT).
While AI systems are nowcapable of generating
text and can check facts, and bots have been used
to editWikipedia in the past, theyaren’t always
the most capable or able to mimiiicccttthhheeetttooonnneeeooofff
humanwriters.
Bytrainingitssystemontwodatabases,
onewithpairsofsentencesandanother
witharelevantWikipediasentennnce,
the researchers at MIT were ableeeto
generateanAIthatcanfindoutdddated
textinWikipediapages,andthennn
rewritethemsimilartohowa
flesh-and-bloodeditorwould.
TheAIsystemworksby
looking at newor updated
information typed intothe
user interface in an unstructureddd
sentence,one without grammar
or style.The AI then finds the
relevant Wikipedia entry,
pinpoints the section the
updated information is
referencing, and then rewrites
the sentence with amended
details in ahuman-like fashion.
TheideaisthattheAIremovestheneedfor
humaneditorstolaboriouslysearchthroughand
amendWikipediaentries.
“There are so many updates constantly needed
to Wikipedia articles. It would be beneficial to
automatically modify exact portions of the articles,
with little to no human intervention,”saysDarsh
Shah, aPhDstudent at the MIT Computer Science
aaannndddAAArrrtttiiififificcciiiaaalllIIInnntelligenceLaboratory,andoneofthe
leadauthorsofapaperdetailingtheresearch.
“Insteeead of hundreds of people working
on modddifyingeachWikipediaarticle,you’ll
onlyneeeedafew,becausethemodelis
helpingordoingitautomatically.Thatoffers
dramaticccimprovementsinefficiency.”
TheAAAIcan also help identify so-called ‘fake
newsss’asithasbeentrainedtorecognise
theeedifferencebetweenalegitimate
WWWikipedia sentence and information
ttthatisfactuallyincorrect.Thesystem
ssseeksevidencetoascertainifsucha
ccclaimistrueorfalse.
Thesystemisnotreadyforfulluse
onWikipediajustyet,asithasbeen
giveeenascoreoffouroutoffivefor
accuracybyhumaneditorsandascoreof
three---and-a-halfoutoffiveforgrammar
accuracccy. As such, it outperforms other AI
tools buuutisn’t quiteable to mimic human
editors just yet. (Phew! –Ed.)
MIT develops AI to edit outdated
Wikipedia articles
NEXT-GENERATIONDRONEScould be powered by
software inspired by howbees adapt to and navigate
their surroundings.
Scientists fromthe University of Sheffield are
reverse-engineering bee brains to createadrone
prototype,which is influenced by theflying insects’
ability to navigateaccurately over several kilometres
and learn environmental features on the fly to then
find their wayback to their hive.
Professor James Marshall fromSheffield
University presented his team’s work at the
American Association forthe Advancement of
Science conference in Seattle,detailing howitaims
to createsmall drones that can effectively navigate
their surroundings as bees do.
“Bees are really consummatevisual
navigators. Theycan navigatea
complex 3Denvironment with
minimal learning very robustly,using
only amillion neurons in acubic
millimetre of brain. Forus, they’re at
asweet spot forbrain size and
intelligence,” Marshall explains.
Despitetheir tiny size,bee brains
can multitask and are able to optimise
the distances the insect flies fromits hive to forage
fornectar, meaningthe brainlearns and adapts to
newscenarios very fast.
Currently,AIsystems used forimage processing
can’tcomputewhat theysee anywhere near as
quickly as some of the smallest natural brains.
To try to replicatehowbees navigate, researchers
have split the project intotwo experiments.
The first involves attaching radar transponders to
bees and analysing their flight paths so researchers
can gain insight intotheir neural processes.
The second experiment involves the more
gruesome process of insertingatethered electrode
intoabee’s brain and then observing its movements
around avirtual reality environment. By
ysing the neural signals, it is hoped
he scientists will gain adeeper look
intobee movements.
“We’ve modelled maybe 25%of
the honeybee brain, maybe atouch
more,” Marshall adds.
“Wehavebee-like robots that
can fly around the lab behaving as
abee would, extracting information
omthe world.”
Bee-inspired software could power
next-generation drones
arounda
anal
th
a
fro
