Artificial Intelligence is a field that’s developing minute by minute and day by day, all over the world. Some of the planet’s greatest scientific minds are working on its development, and we’re told that AI-led initiatives will be the next ‘big thing’ in almost everything, whether it’s personal assistants, self-driving cars, or video games. So why is it that Facebook’s Head of Artificial Intelligence thinks that the development of his specialist field is about to ‘hit a wall?’
The opinion, which comes from the highly-respected Jerome Pesenti, seems strangely timed. Only last week, we found out that artificial intelligence is going to take on a new role in the British casino scene, where it will be used to assist players who use UK online slots. The software will be used to monitor the spending habits of online slots players, and pick up on unusual betting habits. If the AI believes that the players are showing signs of impulse or erratic betting, the online slots will be locked for thirty seconds to give them the chance to reconsider their bet. Gambling wasn’t an area where the potential benefits of AI had been previously considered, and so the news was taken as a sign that the willingness of companies and industries to engage with the possibilities that come with AI was increasing.
Facebook itself has been a major investor and researcher in the field of AI in recent years. Frequently criticized for what’s perceived as a lack of speed and accuracy with alleged instances of abuse on its platform, the social media brand has spent money looking into whether AI could be taught to identify abusive conduct and behavior among its users and clamp down on it immediately. The proposed system is currently at a trial stage, but if successful, it could be implemented to reduce the burden placed on human moderators. The company simply cannot employ enough staff to monitor the activity of its two-and-a-half-billion users at all times, and there have been fears in the past that human operators could be left with trauma and stress from seeing some of the content that the less desirable members of the Facebook community post online. A computer program wouldn’t be prone to any such issues.
Part of the reason that Pesenti believes that the progress of artificial intelligence may soon begin to stagnate is that computing technology is no longer able to keep pace with the rapid progress and increasing sophistication of AI. As was reported by OpenAI last month, the amount of computing power required to run the most advanced forms of AI doubles almost every three months. That increasing demand will soon outstrip the capabilities of even the most advanced machines and processors and will leave the software waiting for better hardware before it can progress to the next level. That creates a problem that isn’t just about machinery – it’s also about cost. Pesenti says that the cost of AI research within Facebook has increased by a factor of ten within the past year and that such sharp rises in price aren’t commercially sustainable.
Where it’s possible to make a profit, there will always be an incentive to spend money. If Facebook believed that it stood a realistic chance of making two billion dollars next year directly from efficiencies or improvements brought about by AI, it would have no reservations about spending one billion dollars to chase that profit. Given their interest in seeing the field progress, they would probably still spend one billion dollars even if they only stood to make one billion in return, and therefore break even. The cold, hard, economic truth is that the potential for profit isn’t there as things stand. Increasingly advanced and clever artificial intelligence could make it easier for Facebook to go about its business, but there isn’t an obvious cost-benefit, and that makes the spending had to justify. Everybody wants to see AI progress, but nobody wants to go bankrupt trying to make it happen.
If a company as large as Facebook, with the enormous financial reserves that it has, can’t see a cost-benefit to artificial intelligence, then it’s reasonable to assume that the majority of other companies interested in investigating its potential will shortly reach the same conclusion. That isn’t to say that AI research projects will grind to a halt, or that ideas will be abandoned, but progress and spending will probably slow down. AI will still make progress, but possibly not at the breakneck pace we’ve been seeing for the past five to ten years.
Pesenti’s assessment of the current situation isn’t completely downbeat. As the head of AI within the company, he still believes that AI-led initiatives can and will make the platform safer for users, and can also help to ensure that the content that users see when they log into the website or app is relevant to them. The latter is an increasingly important point because of Facebook’s reliance on advertising as a means of making money – if AI can help advertisers put their products in front of people who might want to buy them, the company will continue to use it. It wouldn’t be cynical to suggest that ensuring that this happens is a key performance indicator of Pesenti’s job. He also believes that AI entertainment will become more sophisticated, and has spoken of recently seeing a working example of an AI headset being used to allow people to interact with virtual reality objects using their own hands as controllers – the kind of technology which has, in the past, been the preserve of science fiction. It sounds as if Facebook has a lot of artificial intelligence up its sleeve and ready to roll out to users and customers over the course of the next twelve months but may not have anything beyond the projects its currently working on for quite sometime after that. Unless there’s a sudden breakthrough in hardware capability that makes AI development faster and cheaper in the near future, we may all have to hold off on the idea that we’ll all have interactive humanoid robot servants by the mid-2020s.