[Back to Artificial Intelligence]

AI, not the AI Industry

To be clear at the start: this article is about the future of Artificial Intelligence, not the future of the AI industry.  This is partly because I don't know enough about the real details of what is happening in the industry, and that is partly because it is in the interests of the people who run the industry to lie about their achievements and plans.  But we can make some reasonable predictions about the future of the technology.

What we can say about the industry in October 2025 is that most economists who look at it from the outside are expecting the AI bubble to burst fairly soon, and this will presumably lead to a complicated series of mergers and acquisitions, leading to far fewer players in the field.  But this will not change either the actual achievements or the realistic prospects for the technology to any appreciable extent.

A Splintered Market

Nobody is going to find the Holy Grail of AI, sometimes described as 'Artificial General Intelligence', or 'AGI'.  Well, probably not in this century, anyway.

Investors, like most Generals, are always fighting the previous war.  The see the market dominance of a few companies which appeared, sometimes out of nowhere, to control the world of the Internet; they see AI creating a new world, and they want to be running - or investing in - the company which dominates AI the way that Google and Amazon and a few others dominate the Internet.  It's not going to happen.

There is, practically speaking, one Internet - almost by definition.  So if any company is providing tools and services to help us use the Internet, they are automatically competing with every other company providing equivalent tools and services.  Once people decide which search engine is the best, the outcome is pretty clear, and advertisers respond accordingly - giving the winner access to far more funding, and hence making the winner capable of both innovation and research.  The winner doesn't quite take all, but it's pretty close.

But there is not going to be just one market winner in the field of Artificial Intelligence, because there is not just one form of intelligence - and the different forms of intelligence are, to some extent at least, incompatible with each other.  George Monbiot did a nice article about this, if you are in any doubt: There Is More Than One Kind Of Intelligence.  You need one kind of intelligence if you want to build a bridge which will handle the anticipated traffic, and a different kind of intelligence if you want to persuade a community that this bridge will benefit their community; you would not expect an AI which is optimised to argue a case in a law court to also produce an original and eye-catching work of art.  But, in each case, you want to be able to trust that the AI is giving you good advice.

So AI will be used in many different fields, and will need to be optimised for each one.  Sometimes we want creativity, sometimes we want pure facts.  The technology will become part of many different activities: it will act, or advise, or create, or suggest, as is appropriate for each one.  In the foreseeable future, you will not be able to optimise one product for all these situations; in any case, there is no need to.  There can be a large number of successful AI products without anyone getting anywhere near AGI.

We can already see AI products being used in a large number of diverse situations.  For example, they can:

• search the Internet and summarise webpages;
• write articles and essays;
• interpret medical results;
• assist surgeons;
• design scientific experiments;
• write and test computer programs;
• produce special effects for film and TV;
• de-age actors;
• create virtual actors;
• provide a virtual sales assistant, answering your questions about a company or product;
• and we are getting to the point where a virtual personal assistant can answer questions about your diary and warn you about suggest anniversaries and other events you may have forgotten about.

If you want your product to be used in one of these ways, why would you confuse the AI training by including material to enable it to do a completely different task?