sdecoret - stock.adobe.com
The AI market is growing.
Every week, vendors release new tools and technologies that enable enterprises to more easily integrate AI with their missions.
However, as AI technology advances, the industry must confront some serious concerns and challenges.
Vendors and users alike may be able to meet some of those challenges in 2022, while other tests the AI industry must undergo may get more difficult.
In anticipation of a dynamic year ahead , here are eight predictions about significant advances and changes in the AI market in 2022.
1. Regulation and governance
A headline that many think will dominate in 2022 is regulation and data governance.
A shift is underway and will continue next year in how both government agencies and private organizations are approaching AI ethics, according to analyst firm Cognilytica.
Until recently, enterprises in large numbers were acquiring and implementing various AI tools and platforms largely without considering the consequences of biased and unexplainable algorithms.
But by the end of 2022, vendors will not be able to sell major AI systems to a large business or government agency without following specific guidelines relating to bias, transparency, explainability and data set collection.
There's already a framework for ethical AI in the U.S. Department of Defense, noted Ronald Schmelzer, a Cognilytica analyst.
While it's not required, the DOD is asking contractors that want to sell AI projects or tools to the government to show compliance with certain frameworks involving a "human in the loop" development strategy, as well as disclosures about data and any bias in data sets.
"It's not that they're requiring you that you have to have specific things, but they're requiring you to just say what those things are," Schmelzer said.
Specifically for organizations buying AI tools, compliance is also about liability, enabling organizations to shift the blame in case something goes wrong with the technology, Schmelzer said.
Next year is also expected to be when the country starts to move forward on some kind of U.S. version of the GDPR European data privacy law, said Doug Cahill, an analyst at ESG.
"It'll take some political will," he said. "The midterm elections will be interesting in how that will impact such a law in the states."
2. The year of acquisitions and consolidations
During the first nine months of 2021, some $50 billion poured into the funding of AI startups around the world, according to Kashyap Kompella, an analyst at RPA2AI Research.
The same trend applied to acquisitions involving AI vendors, which stepped up this year and are expected to continue into 2022.
Ronald SchmelzerAnalyst, Cognilytica
Some observers also expect notable consolidation in the MLOps sector in 2022.
Amid a wave of mergers and acquisitions involving MLOps technology, vendors may decide to move out of the MLOps area and pivot, according to Kathleen Walch, an analyst at Cognilytica.
Schmelzer said some independent startups focused solely on MLOps will have to make a case for why they are relevant enough to remain independent. Vendors in this category include Seldon and Tecton.
"It's looking pretty much like a foregone conclusion that 2022 is going to be the year you either branch out" or get acquired, Schmelzer said.
3. Short-term investments and slowdown of some AI projects
Other than investment in startups, 2022 will likely also see investments in short-term AI projects.
"The focus is on near-term projects that deliver tangible benefits immediately to help people in this world of hybrid work," Kompella said.
As organizations focus more on short-term projects, some say this could cause a shift from AI – which many enterprises are still finding difficult to implement -- toward automation.
"We're predicting in 2022 there will be a de-staffing of AI projects, and this is going to accelerate," Walch said.
Many enterprises in certain industries such as retail, food supply and customer service do not have the capacity to focus on AI projects at the moment. They must focus on automation because they're short-staffed.
This likely temporary retrenchment involving AI projects will also occur in the government IT sector, Walch said.
Some projects that were in initial phases have recently been reprioritized, she said. Money formerly earmarked for AI projects has been shifted to automation, largely because of tight funding.
This is not a long-term trend and will probably persist only as long as the pandemic does.
4. Virtual assistants and 'the Great Resignation'
The need to fill the employment gap caused by employees leaving jobs during the pandemic is leading to more investment in virtual assistants, or "digital employees."
There will also be investments in several AI-infused providers of Intelligence Assistance and RPA.
However, digital assistants are not always replacing humans, said Dan Miller, an analyst at Opus Research. Instead, digital assistant technology often is a way companies can use AI to do some things people don't want to do.
"Coming out of the lockdown or pandemic, we're finding that jobs are going unfilled," Miller said. "So, you're going to have to figure out what aspects of AI could organically move to do some of those things."
Digital assistants will also help those who have decided to follow other career paths, such as freelance or app-based jobs, Miller said.
"Sometime next year, I think you'll see a series of IPOs or whatever that are around technologies that assist those gig workers," he said.
5. The year of data
Rather than simply turn to automation, enterprises in 2022 will focus on looking at data to figure out whether they need automation.
Organizations should also consider risk as part of an automation project, said Florian Schouten, vice president of product management at AI-based DevOps vendor Digital.ai.
Evaluating risks can include ensuring that enough testing was done or that employees followed company policy during a project.
With analytics technology having reached a mature phase, it's markedly easier now for business analysts and developers to collect data from different places and build their own machine learning models. That trend will continue in the next year, Schouten said.
"The kind of data analysis is moving out of that realm of BI and ETL into your everyday application," he said. "Your typical business analyst is now able to access data much more easily than they were in the past."
6. The year of search and self-supervised machine learning
Next year is also expected to be when search starts to become the predominant entry point for AI-infused customer care, Miller said.
Google pioneered and evolved search. Now more enterprises are deploying more AI technology in the search boxes of their applications, Miller said.
Related to this trend is a move to more discovery of back-end data, Miller continued.
Enterprises have invested billions of dollars into making the information in their back-end systems accessible. They've also invested in knowledge management systems and content management systems.
These investments will have a positive effect on customer satisfaction and make a significant economic impact by sparing organizations the expense of tasking staff or employees with tagging individual conversations with predefined categories, Miller said.
"As an alternative, the AI-infused resources can discover new categories for calls, which often results in speedier resolution of issues," he said, referring to customer service contact centers.
7. The year of a trillion-parameter language machine learning model
This year saw big advances in large language models, and those advances will continue.
Recently, Microsoft revealed that customers will now have access to OpenAI's powerful GPT-3 model with its new Azure OpenAI Service. The tech giant also partnered with Nvidia to release what it claims is one of the largest AI-powered language models: the Megatron-Turing NLG 530B natural language processing program.
In October, independent AI hardware and software vendor SambaNova released its own GPT-AI powered language model with its dataflow-as-a-service product.
Despite criticism that such large language models are environmentally harmful because they require so much energy to power the compute resources necessary to process them, research shows that the models are proving to be powerful and useful, Kompella said.
The models will continue to get bigger and bigger, he added, explaining, "In 2022, we'll see a trillion-parameter model."
8. A pivot from autonomous vehicles to electric vehicles
One area that may see less growth and investment is autonomous vehicle technology.
Many autonomous vehicle enthusiasts have been predicting that fully self-driving cars would be on the road by now. But those predictions have largely been proven wrong, and many organizations and consumers will start losing patience with the pace of development of the technology.
"I don't really know if full self-driving really is going to be possible in the way that people expect," Schmelzer said.
Since more people are starting to see some of the real difficulties and challenges with self-driving vehicles, Walch and said it's likely that multi-billion-dollar investments into the technology will pause.
"I would not be surprised if one of the major auto manufacturers decided to put their full self-driving program on hold," Schmelzer said. "I wouldn't be surprised if one company said, 'you know what? We're going to focus on the bigger market.'"
And that wider market is electric vehicles.