The field of artificial intelligence (AI) is rapidly evolving and creating buzz in nearly every industry. With generative AI entering the scene and opening a range of new AI capabilities, it makes sense that some may have the expectation that AI will continue to grow at this rapid rate. But the thing to remember is that progress in AI is not, and has never been, linear.
In the past, AI research was mostly confined to academic institutions and large tech companies. As AI has democratized, thanks to the availability of powerful computing resources and open-source software, anyone with an internet connection can experiment with AI. This has led to a proliferation of new AI models and applications.
This AI democratization has led to the development of many innovative new applications. But it has also led to some unrealistic expectations around the growth of AI. Progress will not continue exponentially. We’re now entering generative AI’s “tuning phase.” This is a plateau of sorts during which we must examine what we have now to understand how we can improve.
We’re not at the point with generative AI where we understand the extent of its limitations. We’re still experimenting to understand what it does well, where it fails, and how we can tune it to make it do more. We can’t extrapolate what’s happened over the last 12 months to measure where we’ll be even two years from now. It will take time to develop a wider, deeper understanding of the capabilities and limitations of AI algorithms. However, we can still expect to see a series of smaller, incremental breakthroughs that will lead to significant improvements in AI performance.
This is what happened with the development of transformer-based language models, such as GPT-3 and LaMDA. Through slow but then exponential progress, these models were eventually able to achieve state-of-the-art performance on a variety of natural language processing tasks, and they have opened a whole new range of possibilities for AI applications.
In the legal space, certain AI applications need refinement, such as use of generative AI in litigation drafting. Certain other sectors, such as contract review, are applying more established, tried-and-true AI methods that have continued to progress.
It is important to remember that progress in AI is not a sprint but a marathon. It will take momentous time and effort to develop truly intelligent AI systems, but the potential rewards are immense. By understanding the limitations of current AI algorithms and working to overcome them, we can accelerate the pace of progress and bring about a future where AI is used to solve some of our most pressing problems, including inefficiency in legal.
