AI is taking much longer than anticipated to come into its own – here’s why…

14 mins. to read
AI is taking much longer than anticipated to come into its own – here’s why…

Artificial intelligence (AI) is already being deployed in the NHS. But general AI is still some way off. Alexa is not the game-changer that Amazon envisaged. And driverless cars are running late. Victor Hill thinks it’s time to get real on AI.

Slow progress in the NHS

Matt Hancock, the UK Secretary of State for Health and overlord of Britain’s National Health Service (NHS), re-appointed by Mr Johnson when he took over from Mrs May on 24 July, has been a sterling advocate of the need to digitise the NHS. In early August he announced £250 million to fund a new AI lab under the auspices of the NHSX – an agency founded earlier this year to drive the digital transformation of public healthcare in the UK.

Master Investor Magazine

Master Investor Magazine Issue 54

Never miss an issue of Master Investor Magazine – sign-up now for free!

Read the latest Master Investor Magazine

NHSX is already developing software to analyse scans such as mammograms in breast cancer screening. Also in view is a computer-based system to revolutionise the triage system at Accident and Emergency wards. Any British resident who has had to attend A&E with a loved one will know that the ritual is to report to a desk and then be subjected to consultations and tests which determine your place in the queue to see a doctor. New touchscreen devices will take over this function in time and will be able to download vital metrics like pulse and blood pressure as they do so.

More prosaically, NHSX is also investigating systems which would allow citizens to use their smartphones to book and reschedule hospital appointments and to download their test results. Many doctors don’t likely this idea as they consider themselves to be the prime mediators of healthcare – the sole instigators of diagnosis and therapy

On the other hand, doctors’ lives in the UK and elsewhere are being made harder by inadequate IT infrastructure. It appears that different UK hospital trusts use different software packages. (The same is true of different UK police forces – just recall the awful Soham murders (2002) when a known paedophile in Lincolnshire was allowed to become a school caretaker in Cambridgeshire because the two databases were not integrated). Moreover, hospitals work flat out 24/7 while NHS IT support is reportedly unavailable outside of normal working hours.

It’s great that Mr Hancock is so enthusiastic about harnessing AI for the benefit of the NHS, but there is still a lack of ambition in Whitehall in this domain. I have argued before in these pages that people, especially the younger demographic, could upload health data to an expert system almost continuously at zero marginal cost – though they would have to be confident that this data were secure.

New advances

On Wednesday this week scientists at the University of Oxford announced that AI could be used to predict those at risk of a fatal heart attack up to five years in advance[i]. They have developed a biomarker, using machine learning called the fat radiomic profile (FRP). This detects biological red flags in blood vessels supplying blood to the heart, and identifies inflammation, scarring and changes to blood vessels – all pointers to a future heart attack. This approach has huge potential to detect the early signs of heart disease and thus to initiate preventative therapy.

The team compared the scans of 101 people from a pool of 5,487 patients who went on to have a heart attack or stroke within five years of having a cardiac computed tomography angiography (CCTA) to scans from patients who did not have a heart attack. Using machine learning they developed the FRP fingerprint to estimate the level of risk. The more heart scans that are added to the system, the more accurate the predictions will become. The system could be operational within two years.

Gamesmanship – what we can learn from machines

It seems that AI-enabled machines are unsurpassed game-players. But does that really make them intelligent?

Last December DeepMind invited the world champion player of the video game StarCraft II, Grzegorz Komincz, to play against a bot called AlphaStar. Komincz was thrashed 5-nil. The human player later stated that the machine had deployed well-known strategies in extraordinary ways. A previous incarnation of the machine, AlphaGo, beat the legendary Go world champion Lee Sedol in March 2016. This is not an entirely new phenomenon. IBM’s DeepBlue system managed to beat grand master Gary Kasparov at chess back in 1997.

DeepMind is the AI incubator founded by British genius Demis Hassabis and sold to Google in 2014 for $400 million. It has been at the forefront of the development of general artificial intelligence since then. But, recently, DeepMind has been hinting that the point of AI is not to build machines which act intelligently independent of humans, but to enhance human intelligence itself. AI systems and humans will work on problems symbiotically.

Master Investor Magazine

Master Investor Magazine Issue 54

Never miss an issue of Master Investor Magazine – sign-up now for free!

Read the latest Master Investor Magazine

Human gamers who play against AI systems report that they learn from the machines. After losing to AlphaGo, Go champion Fan Hui trained himself by playing against the AI system repeatedly. His global ranking rose from 600 to 300 in just a few months. According to New Scientist all professional chess players now practise against chess computers. These tend to play highly defensively, and as a result the style of top chess players has become more defensive too[ii].

But the very latest generation of chess computers actually play more inventively, eschewing some of the traditional defensive strategies. This is because earlier generations of chess computers contained human biases. The latest generation of machines has improved by playing against each other and are capable of machine learning. This is achieved by building neural networks – layers of software which mimic the capacities of the human brain to learn.

DeepMind’s David Silver thinks that this approach could also be applicable to solving certain fundamental problems in science. AlphaFold is a DeepMind product that has predicted the intricate structure of protein molecules from their amino acid sequence. That makes this expert system a potentially amazing tool in medical research.

But the idea of setting this task is itself a human input. We are still very far from saying that the machine has a mind – and whether that will ever be achievable is the subject of debate. As I suggested in my piece on cyborgs in the March edition of the MI magazine, thinking machines will probably entail some kind of human brain-machine interface.

Alexa! You’re fired!

The market for virtual assistants has gone into reverse. Apple (NASDAQ:AAPL), Amazon (NASDAZQ:AMZN), Alphabet (NAADAQ:GOOG) and Microsoft (NASDAQ:MSFT) all offer virtual assistants which use speech recognition software. All of them as we now know have developed these machines by employing humans to listen in to recordings of their users’ voices. In April it became public that Amazon employs scores of operatives charged with listening to customers talking to Alexa. It seems that Apple does the same with Siri.

Users quite naturally felt that their privacy had been violated. Both firms now face regulatory scrutiny by the EU. And yet consumers’ attitudes to privacy are, to say the least, ambivalent. On the one hand users are happy to post intimate details about their lives and their most trenchant opinions on social media for all to see. On the other hand they cry foul when they learn that their data has been used for commercial purposes. There is nothing new about that. Supermarkets have been using data acquired from loyalty cards to target promotions to their customers for years.

But the fall in grace of virtual assistants goes beyond concerns about privacy. When Amazon launched Alexa the device was hailed as a ground-breaking innovation. And in the last five years Amazon has sold more than 100 million Echo speakers equipped with Alexa. The key is that while Alexa’s functionality has continued to improve, user satisfaction has not.

Apparently, most users ask Alexa to undertake very simple tasks such as checking the weather. If you want to book a flight to Barcelona from the UK it turns out to be much easier to go straight to the EasyJet website and to do it yourself rather than to ask Alexa to do it for you. In fact, “talking” to Alexa and waiting for “her” to respond can be mind-numbingly tedious. And if use of virtual assistants is sub-optimal at home, it has had very little take-up at work.

Driverless cars – in the slow lane

In his new book out this week on AI[iii] the economist Roger Bootle claims that “there is a yawning gap between the hype and the reality [regarding driverless vehicles]” – even though Mr Bootle admits the case for driverless cars is a strong one.

First, Human drivers kill 1.2 million a year and injure something between 20 and 50 million. Some estimates place the cost of road accidents in middle income countries at nearly 2 percent of GDP. Once robot drivers are perfected there shall (so we are told) be no more accidents. Second, a very large number of people cannot drive either because they are infirm or too young and/or do not have access to public transport perhaps because they live outside urban areas. Obviously, the school run which accounts for a significant proportion of total traffic could be revolutionised as children are whisked to school in driverless electric cars. Thirdly, we could use the time travelling in driverless vehicles more productively, for example by reading a book.

Master Investor Magazine

Master Investor Magazine Issue 54

Never miss an issue of Master Investor Magazine – sign-up now for free!

Read the latest Master Investor Magazine

As people prefer to take rides from a floating pool of available vehicles in the fully-fledged sharing economy, so car ownership is likely to fall (as I discussed in this month’s MI magazine). Most cars rest idle most of the time at present; in future a smaller number of vehicles will be in almost constant use, day and night. This could transform urban landscapes as much less space would be required for parking cars. (Most cities are blighted these days not by moving traffic but by parked cars).

In 2012 Sergey Brin, the founder of Google, promised that driverless cars would be available to Google’s employees within a year and that they would be available to all within six years – so, 2018, then. That did not happen. Why?

The main reason is that it has proven more difficult to assure the safety of driverless vehicles than envisaged back then. Despite the claims that driverless cars will be perfectly safe, a 2015 study by the University of Michigan found that driverless cars have more mishaps than human-driven ones. Interestingly, such mishaps are not the fault of the driverless cars. The problem seems to be that, if driverless and human-driven vehicles are permitted to use the same roads at the same time, they find it very difficult to interact with each other. Think about how you change lanes while driving on the motorway. You have to make minute judgments about how other human drivers will respond. It’s much more difficult for human drivers to judge how driverless vehicles will respond. Roger Bootle writes:

The [safety] claims of the manufacturers and developers of autonomous vehicles cannot be taken seriously. For these tests are usually conducted in secret and without independent verification…

In June 2018 a self-driving Uber vehicle killed a pedestrian who was crossing the street in Temple, Arizona. And Arizona has extremely benign climatic conditions for self-driving cars – in Moscow conditions are likely to be much more challenging.

Under new UK legislation drivers of self-driving cars must not take their hands off the steering wheel for more than one minute. But the evidence is that if a car is partially automated then the human driver becomes inattentive and distracted and will be unlikely to intervene when necessary. In 2016 in Florida, the driver of a Tesla (NASDAQ:TSLA) car was operating on Autopilot. The sensors failed to register that a truck was crossing the car’s path. The car steered itself under the truck and the driver was killed.

Bootle thinks that there is little point in autonomous vehicles if a human passenger has to be a “safety driver”. Another point is that if humans rely entirely on self-driving vehicles then they may lose their driving skills altogether. And what if a terrorist organisation could hack into self-driving vehicles and use them as autonomous bombs?

One further point. If self-driving cars were available commercially today they would be phenomenally expensive – not to mention the potential insurance costs. Who, apart from Uber (NYSE:UBER) et al, could afford them? Don’t hold your breath.


This has been a momentous week in British politics. (You have probably read that sentence before.) As of this morning the clouds of unknowing surrounding Britain’s course out of the European Union are darker than ever. As in every good thriller, just when you think the situation cannot get any worse, it does. Mr Johnson is now in office but not in power: he is the prisoner of a Remainer House of Commons in which he now can rely on just 298 votes out of 650. The chlorinated chicken, Mr Corbyn, has given the PM a very nasty (in fact, paralysing) peck.

And the Tory Party is in a state of shock. The cull of 21 dissidents including the Father of the House and five ex-cabinet ministers has been a trauma. One Nation Tories like Damian Green MP are calling for the dissidents to be reinstated; but on the other hand there is a mood amongst the grass roots that it is time to separate the wheat from the chaff.

Master Investor Magazine

Master Investor Magazine Issue 54

Never miss an issue of Master Investor Magazine – sign-up now for free!

Read the latest Master Investor Magazine

Mr Javid’s Autumn Statement prospectively will turn the page on austerity. He will ramp up spending next year by £13.8 billion. The deficit will not rise to more than 2 percent of GDP from the current 1.1 percent. There will be a full budget before the end of the year – but will it be Mr Javid who will deliver it?

The political weather has certainly changed since Mr Johnson entered Number Ten. The Conservative Party is the party of Brexit (Do or Die); and Labour has been re-embodied as the party of Remain. This will allow for much greater clarity in the general election that will surely happen before the end of this year. That election will be the People’s Vote – a chance to re-run the referendum for good or ill.

In Guardian-reading circles the prevailing narrative is that the Brexit mess is all the fault of the unspeakable Tories. But I think history will judge Labour even more harshly. Mr Corbyn has always been ideologically opposed to the EU and sat on his hands during the referendum campaign. The Labour Party voted overwhelmingly in favour of invoking Article 50 and then fought the 2017 general election on a platform of Leave. Now Labour wants a second referendum in which it would campaign to remain. The party’s Damascene conversion has never been coherently explained – not least to the 93 northern Labour constituencies which voted Leave. (NB Great Grimsby.) Messrs Corbyn and McDonnell have called consistently for a general election – but on Wednesday they abstained on the issue.

My best guess is that PM Johnson will be forced to postpone Brexit – much against his will and at great cost to his credibility – probably until 31 January 2020. There will be a general election before that date – probably before Christmas. The key variable is whether the Tories and Mr Farage’s Brexit Party can reach an accommodation – and on what basis. If so, the Tories are likely to win.

How we get to the election is a matter of fevered speculation. There is talk that the Johnson government might seek to amend the Fixed Term Parliaments Act, 2011 so as to remove the requirement that the dissolution of parliament be subject to a two thirds majority in the House of Commons. Presumably, however, the government would require a simple majority to get that measure through – which is by no means guaranteed.

More extraordinarily, the PM could simply declare a National Emergency under the Civil Contingences Act, 2004. This would enable the government to rule by decree – and it could decree an immediate general election. Such a move would almost certainly result in civil disorder and would be very high risk. The markets would take fright.

This week the Bank of England announced that a no-deal Brexit would be “less severe” than previously thought. Meanwhile, Deutsche Bank (ETR:DBK) ruminated that the damage wrought by a Labour government would be less than that pursuant to a no-deal outcome – particularly if Labour were in coalition with the Liberal Democrats who would act as a restraining influence on Labour’s hard left[iv]. The City now believes that a Corbyn government would be manageable.

The pound has actually clawed back some ground in the last few days and is trading early this morning at $1.23/€1.12. The FTSE-100 is upon the week. It is as if the markets have concluded that the crisis will continue indefinitely in a bizarre new normal.


[ii]See: Two minds are better than one, by Douglas Heaven, New Scientist, 24 August 2019, page 38.

[iii]The AI Economy: Work, Wealth and Welfare in the Robot Ageby Robert Bootle.

[iv]See: City warms to Corbyn as sterling tumbles by Tim Wallace and Tom Rees, Daily Telegraph, 04 September 2019. Also Corbyn better than no-deal Brexit, say investment banks, available at:{Paywall}.

Comments (0)

Leave a Reply

Your email address will not be published. Required fields are marked *