Massive language models like OpenAI’s GPT or Google’s LaMDA are generating high expectations. These machine learning models are capable of incorporating impressive amounts of information that they then use to generate statistically probable answers. They are mistaken for a first step towards a general artificial intelligence (AI), capable of generating articulate responses that can be confused with human intelligence and creativity.
In reality these models are very different from the way humans think. We learn language and logical thinking from a tiny set of data when compared to the billions of parameters in GPT 3.5 or LaMDA. But the models cannot distinguish the possible from the impossible, the morally acceptable from the unacceptable, only the most likely answer, even if it is that the world is flat, that there is a superior race or gender. Unlike models, humans are capable of conjecturing creatively, but also of criticizing, of articulating the improbable and impossible, and of moral thinking. Models have to be limited to amoral thinking to avoid controversial responses.
But this article is not about the limitations of artificial intelligence models, not least because others, like my colleague Arlindo de Oliveira, are much more competent to give an opinion on the subject. I recommend reading the recent article by Roberts and Chomsky in New York Times “The False Promise of ChatGPT”.
What motivates me is the area in which I specialize, which is interaction with digital technologies. From this point of view, the services that emerge on top of the models (ChatGPT, Bard, Bing, etc.) are a fascinating object of study, because they represent a potential new disruptive wave. Will AI be a revolution like personal computing in the 1980s, the Internet in the late 1990s, or mobile computing in the 2010s?
A good way to understand the potential is to analyze what each of these waves of innovation represented. Not in terms of the technology per se, but its adoption and economic and social impact. Personal computing dematerialized processes and increased productivity, radically altering administrative, financial and industrial work. The Internet and mobile computing have democratized access to information, revolutionizing the way we communicate and generating new business models, from e-commerce to digital marketing, social networks and the collaborative economy. Each of these waves took five to ten years to reach the maturity of widespread adoption, much longer to have a significant impact on business, work organization and society.
ChatGPT is no different from the canonical applications of previous waves: the spreadsheet of the 80’s, the websites of the 90’s or the mobile applications of the 2010’s. The reason is very simple: adoption is not dictated by technology, but yes by the people and organizations that are slow to adapt. The fact that ChatGPT was the technological service that reached 100 million users the fastest says a lot about how we humans are fascinated by conversational applications (and by the way by speculation), but very little about its effective disruptive potential in the economy and society. society. Just look at what has happened recently with blockchain technology or the metaverse.
The impact of AI models on the services of companies and organizations will be very different from interacting with a chatbot. How Google will be able to incorporate LaMDA into its ecosystem, where it has a monopoly on content and data, will be much more disruptive than Microsoft’s new Bing. If it is confirmed that GPT4 will support multimodality (ie text, image and video), it will imply a radical change in the way we interact with this type of models. The way it is already possible to run Meta’s LLaMA on a personal computer will imply that the technology could be accessible on any device soon.
In all these scenarios the chatbots as an interaction model are not the best alternative.
As the inventor of the personal computer, Alan Kay, said, the best way to predict the future is to invent it. ChatGPT stands for this new wave of innovation as the browser on the first iPhone that made it possible to have the Internet everywhere. It is the first application that allows us to foresee the potential that, in the case of mobile computing, gave rise to AirBnB or Uber. The real disruption of AI will be in those who have the ability to understand that new business models can emerge based on the potential of the models.
In this regard, Portugal is even well positioned, because it has several companies that dominate AI technology and, above all, understand how a successful business can be (Feedzai, unBabel, Sword Health, among many others). Portugal even has one of the largest PRR investments in “responsible AI”, so it is advisable to take advantage of this opportunity. Without forgetting the warning of Kate Crawford, that “AI is neither artificial nor intelligent: it is made of material resources and it is the people who perform the tasks that make it appear autonomous”.
The author writes according to a new spelling agreement