“This time will be different” — will generative AI deliver on its promise?

Heikki Ailisto, VTT Research Professor and head of FCAI’s Industry and Society program, says that generative AI can overcome the limitations of previous generations of AI applications to really create an economic impact. Lue tämä suomeksi

“The widespread adoption of AI will increase the value of production by up to 2 per cent per year for several consecutive years. Finland, along with the US, the UK and Sweden, is one of the countries that will benefit most from AI.” These words could be considered quite fresh in the midst of all the current hype about generative AI. However, they are from a 2016 study by a respected consultancy.

At the time, the emerging trend of deep learning AI was accompanied by strong claims about increasing efficiency and GDP growth, the automation of jobs and the elimination of entire professions from drivers to radiologists.

Over the previous few years, deep neural networks had proven superior to previous machine learning methods in areas such as image recognition—a prime example being face recognition. Deep learning neural networks also proved to be effective in many financial applications and as engines for social media algorithms and advertising. There were plausible reasons to foresee the rapid spread of AI to almost all industrial and service sectors.

What we now know is that, despite the adoption of AI methods and applications in many companies and organisations, the expected huge economic impact has not been seen. There are three main reasons for this.

specificity, effort and competition as the bottlenecks

The first is that AI applications are ultimately quite limited: they can predict certain phenomena and things better than previous methods (for example, the weather or next week's electricity consumption), and they can do pattern recognition: for example, facial recognition or tumour tissue classification. However, these AI solutions are not able to perform complex tasks consisting of different components, such as driving a car in winter conditions, performing an entire assembly task in industry or diagnosing a patient's condition.

Another limitation has been the difficulty of building AI applications: they require the effort of experienced and highly trained specialists.

The third factor is competition: if the new technology is available to all companies in a sector, efficiency gains are passed on to the customer, as prices fall due to competition. Then, by definition, productivity in the sector will not rise, because productivity equals value added per hour worked.

Now, there are similar promises of efficiency and output growth, but also threats of mass unemployment just as there were eight years ago. The reason, of course, is the staggering progress of generative AI. For the first time, we have a machine, or rather software which, on the basis of requests made in natural language, creates a text that seems to make sense on any subject, or even produces images in the desired style—for example, "painted" by a Renaissance master. The latest versions of generative AI understand speech and video and respond with natural speech. It feels like everyone's sci-fi movie visions are now coming true.

So, is this time different from the last? Of the three restrictive factors mentioned above, the one relating to competition is still in force. Instead, the need for deep expertise in building applications will be eliminated when generative AI can be applied even by primary school children. The first constraint—the strictly limited scope of each AI application..will be loosened up, as generative AI is by definition general-purpose.

So, there are good reasons to believe that this wave of generative AI will have a greater impact on society and the economy than the previous one. In education, communication and all other work related to the production of text or image content, the impact of generative AI is already clearly visible.

Many industrial companies and public bodies are actively seeking ways to exploit generative AI. Promising opportunities on the one hand and competitive pressure on the other are keeping organisations looking for new applications.

The large-scale application of generative AI is not even two years old, so it's too early to say how different things will be this time. What is certain is that this wave of AI will touch more people and businesses than any previous AI trends.