Over the past year I have been consulting artificial intelligence models on a multitude of topics, including the fine points of espresso-capable bean grinders, initial versions of legal agreements, how to develop a complex mobile application, the scientific backing for certain nutritional supplements, and the behaviour of flying ants which had unexpectedly swarmed the area I was in. Each time, the AI conversed with me at a level that not only far surpassed the Turing test (of being indistinguishable from a human in blind tests), but that displayed specialist knowledge, understanding of context and advanced reasoning.
This is no surprise: Stanford researchers have found that AI has already surpassed human capabilities on a number of tasks, including natural language inference, visual reasoning and image classification – and is on the way to doing so on other key capabilities.
These recent experiences convinced me that many are misguided in seeing AI as merely a large language model that gives the most statistically likely answers out of millions of options. This is like describing a Van Gogh painting as a series of scribbles that sort of look like the objects he wanted to paint. While technically partly correct, in substance it’s a description that is minimalist to the point of absurdity; and also way off the mark. Similarly, a lot of analysis has gone recently to gauge AI’s impact on productivity, trying to uncover particular percentage ranges of increase within particular amounts of time. Calculations such as these are useful, yet also do not ask the big questions of what AI can ultimately do that will lead to transformational changes in industry and society.
What is not well appreciated is that artificial intelligence agents are just that: agents, not mindless machines. They understand (in a pragmatic sense of the word) what is asked of them, they adopt their assigned roles competently, they take account of context in their analysis, and they reason at a sophisticated level that a mindless machine cannot. In short, they are capable, intelligent entities whose competencies in many thematic areas and types of reasoning outstrip those of smart humans.
The bigger questions for leaders and organisations are not what can be automated or how much productivity will improve, but rather what could be made possible with this kind of unprecedented cognitive and analytical firepower, and what could be the substantive consequences of powerful AI agents for business and society. AI is already reshaping business models and opening up avenues for new ways to deliver customer service and relate to customers. Learning to collaborate with AI to co-create and achieve innovation and transformation is a more effective strategy than thinking of the technology merely as a tool we can use for particular tasks.
Beyond well-publicised concerns about AI posing big risks to humanity, necessitating ethical approaches and careful guard-railing, below are some scenarios that are possible over the next decade, bearing in mind the fast pace of AI development.
In the not too distant future, AI agents will be able to take on certain roles and fulfil particular functions in organisations that are currently seen as the domain of highly trained professionals. These roles and functions will go beyond calculation-based tasks, such as optimising supply chains or keeping financial accounts, towards tasks that require judgement and balancing of a variety of options and constraints. These may include, for example, shortlisting and interviewing job applicants and recommending hires; performance evaluation for different types of roles; developing and implementing marketing campaigns, including selecting key messages and crafting unique selling propositions; and even strategic thinking and recommending corporate-level actions, such as which market segments or geographic markets to target, why and how.
Further, if (or more accurately, when) AI agents acquire legal personality with the attendant legal rights in the same way that companies have, which is a distinct possibility, they will be able to own and control assets and take autonomous actions in the world. Within the context of relevant programming, instructions and goals, they will be able to set up companies and act as entrepreneurs in their own right. AI have already become artists, with an AI painting for example being auctioned by Christie’s in New York for $432,500. In this scenario of AI agents having the status of legal entities, they would be able to hire other AI agents as well as potentially humans. Particular AI agents (or coalitions) could in time acquire economic and social power to a degree greater than that of small countries and could lobby for further recognition and rights. These possibilities raise ethical dilemmas and necessitate careful thinking about what we would like AI to become in our societies. The conditions are present for what would seem like science fiction scenarios a few short years ago to potentially be actualised. But at the same time, the opportunities for innovation and transformation are substantial.
Thinking of AI as merely a tool to optimise organisational processes, improve customer service or raise productivity misses the point. We can both envision and unlock AI’s staggering strategic potential, and be better able to guard against risks, if we recognise AI as capable agents who can perform in many ways as well as, or better than, humans.
Loizos Heracleous is a professor of strategy at Warwick Business School and fellow at the University of Oxford.
This post was originally published on here
