
Artificial Intelligence
The AI Conundrum, Caleb Briggs and Rex Briggs (MIT Press, 2024) is a very good book with the usual flaw which doesn’t diminish its explanation of the technology. I’d recently been sent a book by a publisher and the book was so bad that I couldn’t get more than two chapters into it. I’ll review a book that has problems, but there wasn’t enough to build an article around. Shortly after, I saw this book in the new book section of my local library. I love libraries and this book washed away the bad taste from the other.
The AI Conundrum
The book is aimed squarely at non-technical management who want to better understand the risk and rewards of AI in the business environment. It does that well by beginning with one of the best non-tech explanations of neural networks that underlay current popular solutions. Chapters two and three are very accessible with only a bit of high school math required.
That discussion of math brings up one of the primary issues with the questionable results of large language models (LLM), math itself. The systems are statistical in nature, grabbing information rather than understanding higher math. The problems described with deep learning and math bring up the point that if all you have is a hammer, everything looks like a nail. Even the simplest programs don’t do their own math, they call modules. I’ve written about what are, effectively, expert systems on the front and back end of some, so data can be cleaned up. Neural networks are good, but they’ll only be part of a final system. When the programmers can teach the networks when to call subroutines, it’ll help.
The next chapters provide some nice case studies and define the three cornerstones of precision, input control and rationale (transparency), moving into chapter six’s excellent discussion of business risks.
The second section of the book begins with chapter seven and some case studies. In what seems to be a recurring theme with people who focused exclusively on AI, the historical growth of solutions skips over business intelligence (BI). Much as what is presented as being new with AI had been done for decades with BI, things such as prescriptive and predictive maintenance. Yes, there is some improvement with AI, but what business management needs to do is an ROI analysis to see if the spending is necessary for the results. Refer to my mention of “if you only have a hammer” above and the authors’ discussion of precision within AI which should be extended to precision across technology options, not only between AI models. If non-AI solutions provide good enough results at far lower costs, ignore the fad and expenditure. If the precision of AI is necessary, it’s no longer a fad.
The major problem is one repeatedly mentioned in reviews. They’ve buried the question of job loss in the back third of the book and dismiss it as a real problem. My favorite line is “Assuming they pass on some of these savings to the customer…” There has been nothing in the last couple of decades of large corporations that indicates any support for that assumption. The goal clearly seems to be to keep transferring more money for the other classes to the upper class, as an article on this site pointed out in 2023.
There are going to be major societal changes as AI becomes more pervasive. This is not an industrial revolution; the impact will be far wider. Governments, from cities up through international agreements, are already late in addressing that impact to mitigate the clear harms while supporting the potentially great benefits.
As mentioned, though, that negative is not new to business books about AI. Overall, this is one of the best management books about AI I’ve read in a few years. I strongly recommend it to those people trying to better understand how AI works, the risks and rewards for business based on the current state of the art, case studies showing actual implementations, and a nice framework for business evaluation of AI use.