Artificial Intelligence – Three-Plus Months of Problems and Perceptions, With Hope

As I predicted at the end of last year, AI has found a home in many niches.  But does it seem capable of justifying its $1 trillion economy?

Per “Artificial intelligence is losing hype” on August 19^th, The Economist had concerns – or did it?  The editorial piece’s subtitle, “For some, that is proof that the tech will in time succeed.  Are they right?,” leaves open that AI expectations, especially those not backed up by reasonable data and judicious use of extrapolation, calming down may even predict its broader triumph.  It told us that “according to the latest data from the Census Bureau, only 5.1% of American companies use AI to produce goods and services, down from a high of 5.4% early this year.”  The article compared AI investments with 1800’s British “railway fever,” which, only after it caused an investment bubble, was justified as firms, “using the capital they had raised during the mania, built the track out, connecting Britain from top to bottom and transforming the economy.”  Could that happen with AI?

On September 21^st, the same publication, in “The breakthrough AI needs,” considered what might be required for AI to be comprehensively and gigantically successful, and came up with using more “creativity” to end “resource constraints,” and “from giving ideas and talent the space to flourish at home, not trying to shut down rivals abroad,” as “the AI universe could contain a constellation of models, instead of just a few superstars.”  It is indeed clear that now picking the most successful AI companies of 2050 could be no more accurate than using 1900 information to determine the premier automakers during the industry’s mid-1920s gains.

Most pessimistic of all was “Will A.I. Be a Bust?  A Wall Street Skeptic Rings the Alarm” (Tripp Mickle, The New York Times, September 23^rd).  The doubter, Goldman Sachs stock research head Jim Covello, wrote three months before “that generative artificial intelligence, which can summarize text and write software code, makes so many mistakes that it was questionable whether it would ever reliably solve complex problems.”  The “co-head of the firm’s geopolitical advisory business… urged him to be patient,” resulting in “private bull-and-bear debates” between the two, but the issue, within as well as outside Goldman Sachs, remained partisan and unresolved.

Back to The Economist, where on November 9^th appeared “A nasty case of pilotitis,” subtitled “companies are struggling to scale up generative AI.”  Although, per the piece, “fully 39% of Americans now say they use” AI, the share of companies remained near 5%, many of which appeared “to be suffering from an acute form of pilotitis, dilly-dallying with pilot projects without fully implementing the technology.”  Managements seemed to become “embarrassed if they moved too quickly and damaged their firm’s reputation(s),” and have also been held back by cost, “messy data” needing consolidation, and an AI-skill shortage.  Deloitte research “found that the share of senior executives with a “high” or “very high” level of interest in generative AI had fallen to 63%, down from 74% in the first quarter of the year, suggesting that the “new-technology shine” may be wearing off,” and one CIO’s “boss told him to stop promising 20% productivity improvements unless he was first prepared to cut his own department’s headcount by a fifth.”

Another AI issue, technical instead of organizational, was described in “Big leaps to baby steps” in the November 12^th Insider Today, and started with “OpenAI’s next artificial intelligence model, Orion, reportedly isn’t showing the massive leap in improvement previous versions have enjoyed.”  Company testers said Orion’s improvement was “only moderate and smaller than what users saw going from GPT-3 to GPT-4.”  With high costs and power and data limitations still looming, a shrinking capability exponent could serve to eliminate future releases. 

Six days later, the same source described “A Copilot conundrum,” in that, even one year after its release, Microsoft’s so-named “flagship AI product” has been “coming up short on the big expectations laid out for it.”  An executive there told a Business Insider AI expert “that Copilot offers useful results about 10% of the time.”  Yet the software does have its adherents, including Lumen Technologies’ management forecasting “$50 million in annual savings from its sales team’s use of Copilot.”

An overall problem stemming from the above is that “Businesses still aren’t fully ready for AI, surveys show” (Patrick Kulp, Tech Brew, November 22^nd).  “Indices attempting to gauge how companies have fared at reworking operations around generative AI have been piling up lately – and the verdict is mixed.”  While AI’s shortcomings are real and documentable, many firms “are still organizing their IT infrastructure.”  Reasons mentioned here were “culture and data challenges, as well as a lack of necessary talent and skills,” causing “nearly half of companies” to “report that AI challenges have fallen short of expectations across top priorities.”  So, if AI is overall now a failure, more than its producers are to blame.

A final AI course proposal came from Kai-Fu Lee in Wired.com on November 26^th: “How Do You Get to Artificial General Intelligence:  Think Lighter.”  The idea here was to build “models and apps” that are “purpose-built for commercial applications using leaner models and innovative architecture,” thereby costing “a fraction to train and achieve levels of performance good enough for consumers and enterprises,” instead of making massive, comprehensive large language models which end up costing vastly more per query to use.  It may even be that different apps can use different AI sources which can somehow be combined.  That would be more difficult to organize, but the stakes are high.

This final article points up the thesis of the September 21^st piece above – AI will need creativity in ways less emphasized in the industry.  Companies will need to think outside the boxes they have built and maintained.  There are real opportunities for those doing that best to earn billions or more.  Then, and only then, may artificial intelligence reach its potential.  Designers and executives stopped from exiting through the sides by the massive issues above will need to find ways of escaping through the top or bottom – or through another dimension.  Can they do that?  We will see.