These past two years have ushered in major changes to how we navigate work, human interactions, and information. These shifts have coincided with the increased maturity of AI as a field. As AI has become more widespread, accessible, and acceptable, it’s stepped in to fill gaps in the economic, social, institutional, and political realms – and will continue to do so. With supply chains disrupted, a reduced workforce, and growing demand for social change, we anticipate that this coming year AI will be applied in areas including workplace automation, bias detection, extreme weather predictions, and the identification of COVID-related bottlenecks. Here’s what we expect from AI in 2022. 

Intentions Recognition as a High-Value Business Feature: Intention extraction (i.e., will a particular customer buy, sell, quit, or recommend a product or service) have been available in some natural language processing (NLP) systems for a few years now, but haven’t become one of the standard features that everyone requests. As customer experience (CX) solutions become a bigger piece of a company’s IT landscape, intention recognition will become a must-have feature. Identifying customers that are about to cancel has tremendous ROI to a business, so intentions may become as important as sentiment.

$100 Million Language Model: The race to train the largest possible language model continues unabated, and whether GPT-4 weighs in at a particularly heavy parameter count or another of the tech giants reaches for this particular crown, an organization will announce a transformer-based deep network that costs at least $100 million to train in 2022. Each generation of language models has shown improvements over the previous generation, and with a $100 million model, we could expect to see far greater accuracy in outputs; new behaviors and capabilities enabled by the inclusion of new data sources beyond just text; or even crossing a threshold of intelligence that makes the AI seem profoundly more human, particularly in holding long conversations or otherwise reasoning about extended information sets. However, with inference costs also ballooning with model size, the commercial use case will be limited.

AI Compliance Officers Become Key: With Europe preparing regulations on AI and America potentially not far behind, we’re entering an era of AI regulation for businesses to follow. Especially in America, where patchwork rules will likely show up across state borders, following these rules will be complex and expensive, and a new set of job titles will be added to the corporate lexicon for this new compliance role. While heavily regulated industries like medicine and finance may find any AI-specific regulations not particularly onerous compared to their current regulatory regime, many other industries will have trouble adapting and in particular finding individuals with both the technical chops to make sense of the technology and the legal chops to make sense of the rules. Those who can bridge both skill sets will gain entry to a fast-growing and lucrative career track.

Human Interaction as a Differentiator: COVID and the “Great Resignation” have pushed the trend towards automation of service roles forward, and even if the labor market starts tightening again, we don’t see the effort slowing down. As more store interactions are with automated systems, the human touch will become an important differentiating factor among businesses that embrace it. 

Bias-Detecting AI: Teaching AI to recognize racism, sexism, and other forms of discrimination will become a standard part of the technology toolbox. 2020 and 2021 saw some notable progress in teaching deep learning models to not be discriminatory, but most of this work was in the academic sphere. In 2022, we’ll see this migrate to the corporate arena, where models will be commonly deployed to identify inappropriate content.

Intelligent Clustering Will Start Appearing: Clustering of content around themes and topics has been around for a long time in text analytics, but these mathematically derived clusters often make little sense in practical use cases. The advances in deep learning and available compute power, however, will enable more logical clustering, and we’ll see it finally making its way into commercial products. This will, for example, allow content around flight cancellations related to winter weather to group together, while content around new COVID variant flight cancellations and bans would end up in another cluster.

A New Face of Robotics: Boston Dynamics’ creepy robotic dog has been the face of cutting-edge robotics for a while. However, the robot is a natural place to highlight improvements across all modalities of AI, and we predict somebody will release a friendlier-looking robot that catches all our attention. Possibly, it will be sold as an elder care assistant or an automated store clerk, but using the latest in NLP, voice synthesis, facial recognition, reasoning, and motion control, robotics will really start making science fiction feel real.

The norms around work, doing business, and how we engage with the world are rapidly changing, catalyzed in large part by the ongoing pandemic. With applications ranging from improved social interactions to more efficient workplace practices, AI is primed to meet those changes – and this coming year will be used to bridge labor gaps, streamline supply chains, navigate weather events, and help individuals and organizations navigate the world in a safer, more equitable manner. While the year ahead will no doubt be yet another one of change and uncertainty, we predict that AI will play a major role in how we respond to it.