Despite being a technical field, AI also faces a number of non-technical risks. I would even say that those risks pose a greater threat since they are not specific to a single project or even a single company.
As far as I can tell, the four greatest risks facing AI as a sector are:
Let's review these in detail.
When leaders don't understand what AI is and how it works, they put themselves in a precarious position: they are more likely to pour money into AI and regret their decision later on.
Why so? Because these leaders have misplaced expectations on what it can truly achieve for the business, which skews their view on ROI. They build data science teams that don't deliver, in part because there's nothing for them to deliver. They'll further be surprised when they encounter common failure scenarios. All of this leads to misplaced disappointment, and ultimately contributes to the reports that the vast majority of ML/AI projects never make it to production.
(This lack of leadership-level understanding also skews the data science job market, since their companies overstate their need for ML/AI talent and then claim there is a "data scientist shortage" when they can't fill the (unnecessary) roles. But I'll save that for a separate post.)
A subtle twist on the first risk is when a company has a genuine need for ML/AI, but they mismanage the effort. This leads to low ROI on data work, which ultimately gives the company a bad impression of ML/AI overall.
Consider a company that hires data scientists before figuring out the specific projects that team should tackle. They might luck out, in that the data scientists bridge that strategic gap from "ad-hoc analyses and models" to "embedding machine learning right into the company's business model and product" on their own. But I emphasize both "might" and "luck" here. Most of the time, these situations do not end so well.
Similarly, there are the companies that treat data science as more of a software development effort. A data project is, by definition, an experiment. You can't schedule its date of completion in advance, since there is no guarantee that a model will perform to your satisfaction. Instead, you can only define boundaries of time and effort in order to limit your investment in a given idea.
While many companies have failed to get real traction on their own AI projects, they still eagerly purchase the AI-driven services of others.
You'd think that AI working well would be a benefit, not a drawback. The catch is that, in some fields, AI is not really working as claimed.
Consider adtech, marketing, and other fields that rely on large-scale, sometimes sneaky data collection and sharing. Buyers continue to invest in the adtech ecosystem because they are absolutely convinced of its performance. Even when an adtech provider's own team points out that their product is "almost all crap."
Similar concerns exist in fields such as autonomous vehicles and facial recognition tools. Maybe, some day, these will work as well as the sales team tells you. But as of now, there's still a sizble difference between the present state and that possible future state. (Or, if you'd prefer, the difference between what the technology has already proven it can do, compared to what it might be able to to achieve down the line.) We can't blame people if they see these problems and then question the effectiveness of ML/AI overall.
The three other points culminate into this fourth and final source of risk: missed opportunities. If the initial AI hype boom is the earthquake, this is the aftershock.
Company leaders who continually see ML/AI over-promise and under-deliver will rightfully assume that the field has no real value. So they won't invest. They will therefore miss out on all of the ways AI could actually be of service to their company: providing decision support, automating activities, and adapting to changing conditions (based on new data).
In turn, the companies that need AI but choose to forego it will be limited in their ability to scale. And they will miss out on the counterintuitive-yet-valuable finds that come out of a large-scale, probabilistic view of their data and activities.
Company-wide AI education is key. It starts with the leadership team, as this group of people is responsible for setting the company's overall direection. From there, they must make sure that product leads and other business stakeholders acquire a level of AI understanding appropriate to their role.
Once a company reaches a widespread understanding of the realities of what AI does and how it works, they're ready to plan their data strategy and then map out specific data projects.
Taking this path will dramatically reduce the amount of risk involved in building and deploying AI systems. That should lead to greater return on the AI investment, and wider adoption of AI overall.
Risk Bingo Cards
Adopt a popular expression to improve your risk awareness and assessment.
Not All Datasets Are Created Equal
Some datasets are problems in waiting.