Microsoft & Open AI – Brute force.

Brute force only benefits Microsoft.

• Open AI is making a big bet on compute power being the answer to the limitations of AI in a move that I think will fail but completely explains why Microsoft was willing to throw $1bn at this project.
• Since Microsoft’s purchase of the most expensive lottery ticket in history (see here), Open AI has disclosed that it will spend the proceeds on building a massive supercomputer with the aim to build a human brain-sized model.
• Open AI is of the opinion that “the most benefits will go to whoever has the biggest computer,” meaning that it intends to spend almost all the money it has raised on compute power.
• The deal that it has cut with Microsoft (see here) states that this compute will be built with Azure, meaning that a good portion, if not the majority, of the money invested, will come back to Microsoft in the form of revenues for Azure.
• When this is taken into consideration, one can now understand why a normally prudent company would make such a large and very risky investment in something which is completely unproven.
• Rather than buying a lottery ticket, Microsoft has purchased an option on whether this is the right way to solve the problems of AI and has done so at a price far less than $1bn.
• In effect, Microsoft has purchased a customer and will see a revenue boost whether this approach works or not substantially reducing the risk being taken.
• However, I still think that this is the wrong approach.
• RFM research (see here) has concluded that simply increasing processing power is very unlikely to make algorithms more effective.
• This is because deep learning is experiencing diminishing returns on investment when considering the amount of training compute that is required to derive decreasing increments of algorithm performance.
• Furthermore, increasing compute power does nothing to solve the biggest problem of deep learning which is that the algorithms have no causal understanding of what they are doing.
• This means that they are incapable of taking what they have learned and applying it to a new data set.
• This is the definition of Artificial General Intelligence (AGI) which is the problem that Open AI has clearly stated it is trying to solve.
• Open AI seems to be thinking that if it builds a computer big enough it will have covered enough of the outcomes to render this problem obsolete.
• Furthermore, the Chinese have already built a supercomputer on this scale and as far as I am aware, there is no sign of a real improvement in algorithm intelligence.
• Hence, I remain pretty sceptical about the approach that Open AI is taking, but fortunately for Microsoft, the downside, if it fails, is far less than I had anticipated.
• The outlook for Microsoft remains good but this is increasingly reflected in the valuation and so I am not averse to taking some money off the table.