The way forward for artificial intelligence [Q&A]
There’s been a lot of buzz surrounding the adoption of artificial intelligence. According to a recent McKinsey report 57% of companies now use AI in at least one function. But what about the hype and what has a solid business base?
We spoke to Mike Loukides, VP of Emerging Technology Content at O’Reilly Media and author of O’Reilly Media’s widely cited report AI Adoption in the Enterprise, to discuss the current state of AI and what lies ahead.
BN: Are we moving beyond AI adoption because it’s new and cool to have a serious business case?
ML: We’re currently writing up the results of our own annual AI survey, and what we’re finding is in stark disagreement with McKinsey. About a quarter of our survey respondents have what we call a “mature” AI practice, meaning they have one or more AI applications currently in production. This is much less than 57%. More strikingly, our results have changed little since last year’s survey. About half of those surveyed were evaluating AI solutions – slightly lower than last year’s figures.
There is obviously a lot of flexibility in what these terms mean and how respondents interpret the questions. But that suggests to me that while there is still a lot of experimentation going on, the market has stabilized. It’s less hype, which has shifted to blockchains and NFTs. And going beyond the hype will give AI the opportunity to prove its true worth. There has certainly been a pushback against AI applications that have been wrongfully adopted – for example, CV filtering. And companies are beginning to understand what the real costs are: the costs of recycling outdated models, the cost of data acquisition, and the cost of integrating AI applications into automated deployment pipelines.
But I also think companies are starting to miss out. For example, we’ve seen huge advances in the ability of language models to develop code for programmers. AI will not replace humans; when used appropriately, it will help humans be more efficient.
BN: What, if anything, is holding back the widespread adoption of AI?
ML: In our 2021 AI adoption survey, we found for the first time that demand for data science expertise (including AI and ML) exceeded supply. So there is a talent shortage. One thing that could help fill this talent gap will be the use of AutoML tools to create and train models. We’re seeing signs that these tools are being used more widely in organizations that are newer to AI, as you’d expect.
There is a bigger problem, however. I’ve been saying for a few years that the elephant in the room is getting AI apps out of the developer’s laptop and into production. This year, everyone is saying, “Whoa, there’s an elephant in the room!” We need more people who know how to build data pipelines, test AI software, build deployment pipelines for AI applications, etc. It’s similar to what we call DevOps, or continuous deployment, or Agile, or whatever. But AI applications are not the same as traditional web-based e-commerce applications. They throw a number of curveballs that don’t mesh well with these somewhat older practices. For example, we know a lot about source management with GitHub. But with AI, you need similar management tools for training data. These tools are just appearing. We know a lot about testing. But how do you test applications whose behavior is statistical rather than deterministic?
So, in addition to the AI expertise shortage, there is a second expertise shortage around AI deployment. Call it data engineering, ML engineering, or whatever; it is a significant obstacle. And that requires new or improved tooling.
BN: Will we see an AI-as-a-service model become more mainstream?
ML: Yes. The cloud is an easy way to quickly assemble the computing power you need to train models. AWS, Azure, Google Cloud, and IBM all have attractive tools to help build and train models, and these tools are especially useful for organizations new to AI. We certainly see evidence that these tools are popular among those with less AI experience. What’s most interesting is that Microsoft Azure seems to be in the lead, surpassing AWS.
BN: What role should AI play in the development process? How will developers have to adapt?
ML: I assume you are asking about tools like Copilot from GitHub and AlphaCode from DeepMind. I think they will have a big impact, but maybe not the impact people expect. From talking to people who have used Copilot in production, we’ve learned that it’s not particularly useful for new or inexperienced programmers. He’s not going to “steal programming jobs” or anything like that. But it’s great for making skilled programmers more productive. This allows them to spend more time thinking about how to solve problems and less time searching or trying to remember certain pieces of documentation. This is especially useful for an experienced programmer who suddenly has to dive into an unfamiliar language. And we’ve seen other AI tools to help programmers understand the code other people have written – reading code is a crucial but undervalued skill.
BN: Can we expect to see new positions created to meet the needs of AI?
ML: Certainly. Training and retraining models will become a specialization in their own right, separate from AI programming. Collecting and documenting data appropriately will also become a new professional role, along with other roles associated with data governance. Although many organizations have not yet realized it, we are long past the point where you can use all available data, regardless of how it was collected, its impact on privacy and the inherent biases of data itself or to the data collection process. I have already mentioned AI operations, which will need to consider the differences between AI applications and the business applications we are used to running. Although automation will play a part in all of these roles, humans will be needed to set the direction. I don’t think automation will eliminate any of these roles. Given the scale of modern data problems, I believe automation will make these tasks possible.