As Generative AI is attracting lots of attention, there is much hype around it. Till today I have overheard at least a couple of consultants to consult clients to use Generative AI for anything from customer-service chatbots to better demand forecast. While Gen AI is great for certain tasks such as synthesizing text, it should not be an automatic choice for everything.
As of today, Gen AI is trained to generate text and images, its strengths are in information retrieval, content creation, language understanding and generation. These strengths make Gen AI most suitable for automation such as customer-service chatbots, code co-pilots or personal assistants.
To make better forecast or derive business decisions from data, Gen AI is not the most suitable technology. There have been studies showing LLMs are as good as a random guesser in Causal Inference (estimate causal effect from data). The reasoning ability of LLMs is also highly questionable. Though minor tasks can be automated with Gen AI, the accuracy of AI applications in more complex real-world scenarios cannot be relied upon.
Meanwhile Causal AI, an emerging field of AI, is designed to solve these problems by incorporating cause-effect reasoning. The technology helps you uncover causal relationships and estimate causal effects from observational data with AI. This enables you to make better business decisions with AI-assisted data analytics and causal predictions. An example is allowing ridesharing companies to predict customer demand more accurately by understanding and modelling the causal relationship between different factors, such as weather, festivals and promotional campaigns.
However there is an area where Generative AI may have an advantage over Causal AI in decision intelligence is simulation. An example is Wayve, a self-driving car startup in the UK, has used Gen AI to generate simulations of future scenarios conditioned by natural-language instructions. If one has a lot of data and compute, Gen AI can be potentially trained to generate highly-complex predictive scenarios.
In conclusion, while Gen AI should not be the automatic choice for every use case, it has its strengths in certain tasks such as synthesis of text, images, automation and simulation. Even though Gen AI excels in these tasks, for tasks such as forecasting and decision-making, Causal AI is likely to be the more suitable and accurate choice.
