Blog post

Generative AI could be the trigger for the Productivity Wave we Need

By Andrew White | April 07, 2023 | 0 Comments

ProductivityGenerative AIArtificial IntelligenceChatGPT

AI’s False Dawn?

This was the title of a blog of mine from a couple of years ago: 2018: Won’t See a Massive Productivity Boost From AI – 2019 Might Show It.  My blog at the time tried to synthesize a story from some recent (at the time) news articles:

I had mused in late 2017 the following: Raising Productivity is Our Number One Task.  Frankly, productivity in most advanced nations has been slowing down or measly for many years.  See Innovation and the transatlantic productivity slowdown: A comparative analysis of R&D and patenting trends in Japan, Germany, and the United States (Brookings, 2020).  Even the “catch-up” growth in productivity in emerging markets has also flattened off.

It is productivity that provides economic growth.   Our working populations are contract and aging at the same time.  Fewer folks will pay taxes at the same time as healthcare demand will skyrocket.  If we don’t grow productivity we will have big issues.  The last 30 years were challenging economically.  The next 03 will be of a very different scale.

Its all about Productivity

Working in the IT (information and technology) industry I am surrounded by folks who believe, as I do, that information and technology is at the heart of digital business ambition, growth and productivity improvement.  But we have all been waiting for a renaissance in productivity.  There are many studies on this issue. See Understanding and Addressing the Modern Productivity Paradox.

While I was sanguine in 2017 and 2018, and wrong in 2019, I think time the coming.  Earlier this year I was still not excited: The Death of Innovation (November 2022).  But the more I read about generative AI, the more I think that this specific wave of AI is different.  I get the feeling that it is sufficient to shift our innovations from a more general purpose technology toward something like specific purpose technology.  See Where You Spend Your Firms’ Capital Matters.

Here is a new example that caught my eye in the press yesterday: Expedia launches in-app ChatGPT travel planning feature.

Not All Data is Equal

Let me explore the Expedia example, and even compare it to what Gartner does in its research business.  One argument is that tools like ChatGPT could makes numerous holiday agents (and Gartner analysts) redundant at a stroke.   From a productivity angle this is a good thing.  Costs may go up short-term due to development and implementation.  Long-term, all other things being equal, costs should go down and ability to serve clients – demand – is expanded.  Looking at Gartner’s’ business, it looks similar.  If analysts answer questions based on synthesis of research, ChatGPT could do something similar.  Instead of asking for help to find a suitable hotel or beach, you might search for a suitable best-practice for cloud migration.  ChatGPT could be used to answer those same questions.  Productivity would be improve everywhere.

But there is another side to this conversation.  As with any other innovation that threatens to disrupt work and employment, workers may get displaced (redundant) but also new opportunities will emerge (reinvent).  Often times more technical roles emerge. Expedia would train its ChatGPT tool on all the public data on hotels, holiday destinations, and all the services on offer that go with all of that.  Expedia assumes owners of those places want to be found, so they push their data out and keep it up to date.  ChatGPT would be re-trained all the time.  So Expedia is a great example of a good-fit use-case for.

Some Data is better than other Data

Additionally it is apolitical.  Few folks would select a holiday location because of political leanings.  As a result, there probably won’t be much of a backlash in the data used by Expedia; it seems a stretch that bias would be a problem.  I guess I could say never say never.  Unfortunately this is not what you get if you consider CNN or Fox.  If those organizations were to use ChatGPT, presumably they would use on their curated data.  As such ChatGPT would offer predictably biased responses.  It could still improve productivity for those organizations however.  So bias will not prevent that.

Gartner’s’ business is not quite the same as Expedia here.  Gartner’s body of knowledge is large but private (behind the paywall), and for the most part, changes slowly over time.  Periodically there are parts of the corpus that change quickly, but it is a vast network of interconnected insight.  That insight, if it is any good, is not public knowledge.  The point is we can sell our unique insight for a premium.  So other than being non-public, it looks similar to Expedia.  But there is another big difference.

Data is Everything

In the Expedia example the publishers of the data being used by ChatGPT are motivated to keep their data up to date.  By design they need to – else the purveyors of holiday locations won’t compete effectively.  For Gartner, the data could comprise case studies, analysis, ruminations, and insight, all developed by experts.  But the advice that comes from that data is not a simple, cold, analysis of the data.  There are always leaps and jumps.  Analysts always look for something odd, or different.  Analysts also do pattern recognition.  But the best analysts go out on a limb.  They look for unconventional connections.  They seek something anomalous in order to find the next leap.  ChatGPT is not designed to do this.  It does not create; it is designed to synthesize.

At the same time analysts look through the responses and data we collect from the market.  I cannot tell you how often reference calls with end-users of software are far more telling from what is not communicated, than what is.  ChatGPT can’t really synthesize what is not said, at least not right now.  So ChatGPT cannot really replace all that an analyst does.  But it might be able to operate like the Expedia example and help drive productivity of the more common client questions, or those that assume a stable, curated data set.  And the smartest analysts might help curate the sets for the specific use-cases clients care about.

Data is not quite Everything

One last thought (updated post publication).  My favorite AI colleagues at Gartner, upon reading this blog, shared a last thought.  Data is not quite everything.  With everything going on, it should be clear by now that unless these generative AI models are not governed adequately, the uses of them might be risky..  So data might be most of everything, but without people to help govern the output, we run big risks.

Of course everyone is trying to figure out how to use ChatGPT and generative AI.  The more I read about it, I do believe it could signal a major change in the accounted for contribution of AI to productivity.  What I thought might have taken place in 2019 may well emerge in 2024.   Others have written about the lag in innovation investment and the time it takes for a business impact to be recognized and accounted for.  The investment in computing didn’t drive productivity improvement for many years.  This is true of electricity.  It is true of AI.  See Innovation-productivity paradox (OECD, 2021).  In each case, complementary and dependent investments and innovation had to take place in order for the impact to take place.  Sometimes it is also triggered by increased in training and skills employed by workers to learn how to use the new technology.

ChatGPT and generative AI could be the next big thing to unleash the innovation and productivity we sore badly need.

The Gartner Blog Network provides an opportunity for Gartner analysts to test ideas and move research forward. Because the content posted by Gartner analysts on this site does not undergo our standard editorial review, all comments or opinions expressed hereunder are those of the individual contributors and do not represent the views of Gartner, Inc. or its management.

Comments are closed