Loading...
Tech Trend

Getting Smart About Five 2022 AI And Machine Learning Predictions

Director of Marketing at Appen, driving responsible, inclusive AI and training data conversations with global companies.

As new technologies exit the realm of dreams and enter the realm of possibility and reality, there’s always a reckoning with their true capabilities. In 2021, I saw teams build the foundations for some really important AI changes in 2022 and to turn over a new leaf of more responsible, efficient and cost-effective practices in AI tech.

For communicators in the AI space, this means we should be mindful of the types of expectations we set for 2022. After all, happiness is the difference between expectations and reality.

Responsible AI: The Shift From Aspiration To Requirement

Responsible AI was one of the hottest topics in 2021. While everyone was talking about how to make AI more responsible, I saw very little adoption and change in this area. Our Appen 2021 State of AI report (via Business Wire) found that concern over AI ethics remained at just 41% among technologists and 33% among business leaders. 2022 is the year that businesses should realize responsible AI isn’t just for optics — it could actually lead to better business outcomes.

As companies begin to prioritize responsible AI, so might governments. Governments may start to worry about irresponsible AI. AI could soon follow the journey of data privacy from minor concern to regulation. Gartner predicts that by 2023, all personnel hired to work on AI development and training will need to demonstrate expertise in responsible AI.

Our job as communicators is to make sure our organizations understand the importance of responsible AI and to amplify the efforts in this area as they deploy them, as well as to act as gatekeepers for high-risk releases.

Data’s Key Role In The Entire AI Life Cycle

While many businesses have matured their AI programs, data science and ability to develop machine learning models, some are still struggling with data. They’re beginning to recognize the need for data at multiple stages of AI project development, and some are going outside their organizations to expand their capabilities.

The AI life cycle requires data at a number of key junctions along the way to deployment. Businesses often struggle to manage the many facets of data within their projects. They must acquire, prepare, evaluate and manage data throughout the project. Pair that with the fact that many businesses now need more data faster — which they can often only gain through automation — and it’s no wonder that they’re turning to outside sources for data.

In 2022, I expect that bringing data management for AI more to the center of the conversation and addressing the data for the AI life cycle will not only be a great way to achieve results but also a good measure through which to build trust in AI systems.

Rise Of Synthetic Data

One of the major struggles I saw for businesses in 2021 was data acquisition and management, so it’s only natural that they would look for solutions in fields like synthetic data generation.

Synthetic data is data created by an algorithm, not collected from the real world, and it can come with a lot of privacy and security benefits. By 2025, Gartner (via ZDNet) expects generative AI, which can generate synthetic data, to account for 10% of all data produced. That’s up from 1% today.

Companies like Apollo are already creating synthetic datasets for use with autonomous vehicles. Other emerging examples of synthetic data use can be found in health care, as well as in the more familiar field of marketing, where generative AI helps with marketing content identification and personalization, like in the case of personalized synthetic ads. As implementation and experimentation with synthetic data progress, I think we will see more use cases emerge and adoption pick up.

Advancement Of Internal AI Use Cases

Seventy-four percent of the State of AI respondents said they have an AI budget of over $500,000, while the number of companies with budgets between $500,000 and $5 million has increased 55% year over year. Our report found that those with budgets over $1 million are more likely to see AI projects reach deployment and deliver positive ROI.

As returns grow, I believe the winners in the space will be the ones using AI to solve smaller-scope problems at scale, as reflected in the fact that we found the number one AI use case (62% of respondents) is supporting internal operations.

This move toward internal-facing AI use cases will likely continue into 2022 as teams figure out how to best move data throughout their organizations — by deploying new technology platforms to eliminate silos or by deploying strategies to manage data throughout the AI life cycle.

While this isn’t as exciting as automating a process end to end with the help of AI, as communicators, we should find new ways to bring AI into the story without revealing the proprietary components of the inner workings of our organizations.

AI Checks And Balances In The Mainstream

Machine learning models are dynamic — organizations shouldn’t deploy and forget them. In 2022, I believe the need for continuous model tuning will become critical for most AI use cases. When you are running AI models in dynamic environments, you should regularly update and retrain them to remain relevant as new data comes in.

In 2019, Gartner found that almost 40% of companies have already implemented AI in some form. Given the maturation of the AI industry, I expect that enterprises will shift their focus in 2022 from implementation to optimization. This could result in an increased reliance on model evaluation, tuning, solutions and partnering with vendors that can assist in this process.

When I look back on AI in 2021, I see that it was a time of discovery and discussion. Foundations were built, and now we should set expectations for an ethical, efficient and useful future for AI in 2022.

Source:Getting Smart About Five 2022 AI And Machine Learning Predictions

Leave a Reply

Your email address will not be published. Required fields are marked *