Despite its enormous promise, the science of artificial intelligence has been somewhat of a backwater in the financial world. There are companies that have ridden the AI wave in significant ways: Google claims to have refined many of its services with the help of AI, machine learning has boosted sales of Nvidia’s graphics processing units, and TikTok’s algorithm is said to be a big part of what keeps users coming back to its short videos.
However, it is difficult to identify a pure AI firm that has emerged on the strength of the technology, or a large new market that has been formed. That picture may be poised to shift dramatically.
Generative systems, which generate text and graphics automatically from basic text prompts, have improved to the point that they may have wide-ranging commercial applications. A partner at another top Silicon Valley venture capital company, who sees AI’s previous past as a cemetery for start-up investors, also claims that the race is on to identify breakthrough uses for this new technology.
Since the release of OpenAI’s GPT3 text-writing system two years ago, generative models like these have been all the rage in AI. The ethical challenges they bring are substantial, ranging from any biases they may pick up from the data they are trained on to the potential that they may be used to spread falsehoods. However, this has not stopped the search for practical applications.
Three things have happened to transform these systems from smart party tricks into potentially helpful tools.
One example is that AI systems have progressed beyond text. Meta presented the first system capable of creating a movie from a text or image cue last week. That breakthrough was considered to be two years or more distant. Not to be outdone, Google reacted with not one, but two AI video systems of its own.
The largest AI advance this year has been in image generation, owing to systems like OpenAI’s Dall-E 2, Google’s Imagen, and the start-up Midjourney. Images will be the “killer app” for this new kind of AI. According to him, this new creative tool might be transformative for the generation who grew up with TikTok and Snapchat. It also poses an apparent danger to anybody whose career is based on the creation of pictures in other ways.
The second significant development is the fast decreasing cost of training massive AI models. Microsoft’s $1 billion investment in OpenAI three years ago emphasized the exorbitant cost of this for ever-larger models. New approaches that provide high-quality outcomes by training neural networks with fewer layers of artificial neurons are altering the picture. According to Mostaque, the computational resources utilized to train Stable Diffusion would have cost less than $600,000 at market pricing.
The third development has been the availability of technology. Google and OpenAI have been leery about making their technology broadly accessible, partially due to concerns about potential exploitation. Midjourney’s image system, on the other hand, is open to all customers under a freemium payment model. Stable Diffusion has gone even farther, open-sourcing its software and disclosing information about how it trained its system. This allows other organizations to train an image model on their own data sets.
The hazards posed by such generative systems have attracted a lot of attention. They generate new pictures or text based on the millions of instances from which they have learned, with no knowledge of the original content. This may lead to illogical outcomes as well as purposeful misrepresentation.
However, in a corporate situation, at least some of these flaws may be mitigated. The challenge will be to discover methods to integrate technology into traditional work processes, developing tools that can propose new ideas or speed up creative creation while human employees filter the output. The concept is already being utilized to produce computer code.
According to one investor, the key issue today is whether established industry titans such as marketing, media, and entertainment will be the first to adopt these powerful new creative tools. Or will they be disrupted by a new generation of upstarts with AI roots?