The Daily Telegraph - 23.08.2019

Boss’s absence creates a headache for DeepMind

D

eepMind has been hailed

as one of the greatest

British success stories of

the modern age. When

Google announced it had

bought the London-based

artificial intelligence lab in 2014 for

£400m, few people had heard of the

company or its founders, Demis

Hassabis and Mustafa Suleyman.

Today, DeepMind is supposedly on

the way to creating an “artificial general

intelligence” that will work like the

human mind does, and is believed to be

one of the most prestigious AI

companies on Earth.

But the road to success hasn’t been

easy. This week it emerged that its

enigmatic co-founder Suleyman has

been placed on leave for reasons that

remain unknown.

Suleyman led the charge to turn

DeepMind’s research into products that

could be integrated into Google’s

products, helping it to reduce its data

centre cooling bill by 40pc and

improve its Google Translate service.

A spokesman for the firm said that

Suleyman is “taking time out right now

after 10 hectic years”.

It comes after news that debts are

mounting. According to its latest

accounts, the Alphabet subsidiary saw

its losses climb 55pc to £470.2m in

2018, taking its cumulative losses to

more than £1bn. The deepening debt,

which is due to be paid back to its

parent company by October, will be

covered by Google in an arrangement

that illustrates how critical DeepMind’s

work is to the search engine.

With Suleyman on leave and the

business’s losses growing, it would

seem the pressure on DeepMind to

make money is higher than ever before.

DeepMind’s algorithms have proved to

be game changing in AI. But the latest

accounts raise a vital question: is

trouble brewing at DeepMind?

Many academics and investors have

been sanguine about DeepMind’s

ability to continue running at a loss.

Walter Price, manager at Allianz

Technology Trust, a UK-listed fund,

believes DeepMind has “no need to

make a profit” at all. “Google considers

it to be an R&D effort,” he says.

DeepMind has effectively become a

research wing of Google that the

company has been happy to pour

system in humans – proved its mettle in

2016 after sweeping away the world’s

best player of Chinese board game Go.

For Hassabis, the possibilities extend

into the medical field. But taking its AI

to new sectors will inevitably mean

more investment. Its latest accounts

showed revenues doubling to £102.8m

last year, though this has not been

sufficient to offset rising staff costs,

which, when combined with other

administrative expenses, totalled

£568.3m in 2018, up 70pc.

DeepMind’s main revenues come

from sales of its code to Alphabet,

allowing divisions such as Waymo, a

driverless car unit, to make use of its AI.

No company, even Alphabet, which has

seen its cash pile rise to $117bn (£95bn)

recently, has an unlimited resource of

money into, especially if it helps save

money elsewhere.

“It might be advantageous from a tax

basis to treat it as loss making to offset

the profits that Google is making in

Europe or the UK,” Price says.

He also points to DeepMind’s claim it

reduces the energy consumption of

Google data centres, which may “offset

the cost of the operation” at DeepMind

by making savings in other places.

Given DeepMind’s track record, the

value it offers to its parent company has

become evident as a string of

achievements helped justify the heavy

spend on its army of AI researchers.

AlphaGo, the company’s machine-

learning system that harnesses neural

networks – algorithms designed to

mimic the pathways of the nervous

Building human-like

AI is no easy task, but

with its founder on

leave the goal looks

more illusive, find

Hasan Chowdhury

and James Cook

Technology Intelligence

More activist

than

entrepreneur,

Mustafa

Suleyman’s route

to DeepMind was

far from usual.

He dropped

out of Oxford

University at 19

to help set up a

counselling

service called

the Muslim

Youth Helpline

before founding

DeepMind with

childhood friend

Demis Hassabis

a few years later.

While

Hassabis, a

childhood chess

prodigy, was

regarded as the

academic

figurehead of

DeepMind,

Suleyman was in

charge of

finding

real-world uses

for its

technology. He

was involved in

one of

DeepMind’s

flagship health

projects, the

Streams app,

developed with

London’s Royal

Free Hospital.

Two years ago,

the deal was

found to have

transferred

patient data

illegally.

Suleyman

handled the

fallout. It is not

clear what has

triggered his

leave of absence,

which

DeepMind says

was mutually

agreed and

will see him

return before

the end of the

year. “Mustafa’s

taking some

time out right

now after 10

hectic years,” a

spokesman said.

Mustafa Suleyman DeepMind’s big thinker

capital. So if a day came when

DeepMind had to start making money,

what are its options?

One route to profitability could come

through its new health ventures.

Protein folding, the process by which

chains of protein building blocks fold

over each other to form 3D structures,

has become a point of obsession for

Hassabis – and for good reason.

By improving the ability to predict

the structure of a protein, researchers

can get better at developing drugs to

target proteins that misfold and end up

with an entirely different structure

from one previously seen.

Diseases such as Alzheimer’s,

Parkinson’s and Huntington’s, are

believed to be caused by the misfolding

of a protein. If scientists can get better

at predicting the structures of proteins

that cause these diseases, the more

effective drug development can be.

Turning DeepMind’s new protein

folding ability into a source of revenue

is somewhat of a mixed bag for the

time, however.

“There quite clearly are major

implications” from developments in

protein folding, says Prof Michael

Sternberg from Imperial College. But

there are alternative options to

DeepMind’s system, many of which

have been released for free online.

“What DeepMind is doing, it’s better,

but it’s not unreproducible by other

people,” he says.

The company has other healthcare

applications for its AI that could be

used to generate meaningful revenue.

Last year, its subsidiary DeepMind

Health was transferred from

DeepMind’s control to Google.

DeepMind’s “Streams” app brings

together health data into one app that

helps doctors and nurses track the

health of hospital patients.

The app has been tested at the Royal

Free Hospital in London since 2016,

and DeepMind plans to turn on the use

of AI inside Streams to help hospitals

automatically track patients.

Handing over this app to Google

made sense, says Prof Paul Leeson from

the Cardiovascular Clinical Research

Facility at the University of Oxford, as

the tech moves from DeepMind’s world

of academia to Google’s domain of

developing products.

“The technology about how you

develop an app, how you get it on to

people’s phones, this is the kind of

thing Google is very good at,” he says.

But transferring Streams to Google

doesn’t mean that DeepMind has solved

the challenge of turning its AI

algorithms into moneymaking services.

For DeepMind to begin to generate

meaningful revenue for its parent

company, it will need to demonstrate

that its tech can’t be replicated.

With one DeepMind co-founder

currently taking time away from the

company, that task is looking harder

than ever.

The Daily Telegraph Friday 23 August 2019 *** 35

RELEASED BY "What's News" VK.COM/WSNWS TELEGRAM: t.me/whatsnws