How to avoid AI pitfalls in business
The use of AI in business is still in its infancy, but is already raising fundamental questions for businesses and their customers. While we’re figuring out answers, knowing how to avoid major AI business failure is arguably as important as knowing how to build great AI technology.
Many AI project pitfalls are technical in nature, and businesspeople have limited influence. But some relate directly to business aspects of AI work, and business stakeholders can make a big difference.
Below are five business-oriented challenges for AI work, and tips to avoid or handle them. They’re a result of anecdotal, personal and published experiences of AI projects that could have gone better.
Poorly Thought-Through Business Questions
An easy AI trap for the unwary is expecting AI to answer unreasonable or unrealistic business questions. This can lead to answers that are at best difficult to use constructively, or worse, meaningless or ambiguous.
For example if you have a customer retention problem, you might ask for AI that spots customers likely to leave. It seems sensible to assume that if you know who they are, you’re a step closer to doing something about it. The pitfall comes when you’re asked how accurately you want to identify them, because an obvious answer could be “as accurately as possible”, ideally 100%.
This is a problem if you’re in an industry where customers leave because of growing dissatisfaction over time (rather than individual triggers like price rises). That’s because the most accurate prediction of a decision to leave will be possible when a customer is most dissatisfied, typically just before they make the decision. By then, options to change their mind are likely to be fewer and harder than earlier in the relationship.
A better business question could have been asking AI to uncover early signs of dissatisfaction, ideally with some insight into the causes. With that kind of information, interventions are probably more likely to work.
A straightforward way to prevent this kind of problem is playing out how the AI results will be used in business terms. It’s also very good practice to extrapolate this a few quarters and years into the future, trying to visualise the longer term business impact of this and related AI solutions.
Unquantified or Unstated Business Goals
There’s plenty of material around describing how AI can benefit business, but much less on how to quantify that benefit. Without benchmarks and published data to help validate predicted returns on AI investments, business cases are likely to be riskier than other technology spend. But when everyone seems to be talking about and using AI, it can feel risky to not be doing AI work.
The pitfall for business leaders is starting AI work without clarity on what it’s expected to achieve, expressed in measurable terms. It may seem unlikely that businesses would do that, but anecdotal evidence suggests it’s surprisingly common. The reasons are complex, and the resolutions are challenging.
One possible solution is grouping AI spend with existing investments such as R&D. Another is setting “soft” business goals, consciously accepting latitude on results - including occasional failures. A third is using early projects to create your own baseline of metrics for future AI business cases.
Insufficient Focus on Business Impact
Like other IT work, AI business projects create technical outputs to help businesspeople solve business problems or improve business activities. And as with other IT work, solving problems or improving activities involves modifying or even transforming day-to-day work to make use of the technical outputs.
Today, good IT projects balance business and technology activities, but that’s not yet the norm in AI work. It’s not clear why, but it may be linked to AI being seen as a very specialist technical skill, and business people not being considered qualified to be involved.
At this early stage in the evolution of AI business work, one answer is to simply make sure business implementation is a standard part of any status reporting, including in particular for senior management. Another is to include business people in core AI teams, not just consult them occasionally for “user” input. This is part of a much bigger discussion, and we’re not even scratching the surface here.
Hiring the Wrong Skills
If you staff your AI projects in-house, you’ll need to hire AI skills such as data science. Alternatively, to use third parties for AI work, you’ll need a different profile of in-house AI skills e.g. to procure and manage data scientists or AI consultancies. A common pitfall here is hiring AI skills without being clear on what those should be and why.
It’s made harder by lack of consensus on how to structure AI teams or even where in an organisation they fit. Also, HR and procurement teams are rarely familiar with how to evaluate AI candidates or suppliers.
There’s no simple mitigation here either, but being aware should make it easier to avoid expensive mistakes, particularly around permanent hires. A useful rule of thumb might be to err on the side of flexibility until you have at least a clear AI sourcing strategy.
AI For the Sake of AI
The fifth AI pitfall is the easiest to spot and deal with. It’s the risk of falling into an “Emperor’s New Clothes” mindset, doing AI work primarily because of a desire to use AI, perhaps because everyone else seems to be. This goes hand-in-hand with insufficient thought on the kind of problem to solve with AI, and leads to projects lacking business focus, or trying to solve problems not suited to AI.
To avoid this, look for symptoms that a desire to use AI is driving a project, rather than a genuine, appropriate business problem.
Like surgery and soccer, success is the ultimate goal of AI projects, but avoiding failure can be as important - moreso sometimes. Avoiding these five and similar pitfalls puts you ahead of many others exploring AI in business.
By Was Rahman, CEO of AI Prescience and the author of AI and Machine Learning
Facebook Develops AI to Crackdown on Deepfakes
In light of the large tidal wave of increasingly believable deepfake images and videos that have been hitting the feeds of every major social media and news outlet in recent years, global organisations have started to consider the risk factor behind them. While the majority of deepfakes are created purely for amusement, their increasing sophistication is leading to a very simple question: What happens when a deepfake is produced not for amusement, but for malicious intent on a grander scale?
Yesterday, Facebook revealed that it was also concerned by that very question and that it had decided to take a stand against deepfakes. In partnership with Michigan State University, the social media giant presented “a research method of detecting and attributing deepfakes that relies on reverse engineering from a single AI-generated image to the generative model used to produce it.”
The promise is that Facebook’s method will facilitate deepfake detection and tracing in real-world settings, where the deepfake image itself is often the only information detectors have to work with.
Why Reverse Engineering?
Right now, researchers identify deepfakes through two primary methods: detection, which distinguishes between real and deepfake images, and image attribution, which identifies whether the image was generated using one of the AI’s training models. But generative photo techniques have advanced in scale and sophistication over the past few years, and the old strategies are no longer sufficient.
First, there are only so many images presented in AI training. If the deepfake was generated by an unknown, alternative model, even artificial intelligence won’t be able to spot it—at least, until now. Reverse engineering, common practice in machine learning (ML), can uncover unique patterns left by the generating model, regardless of whether it was included in the AI’s training set. This helps discover coordinated deepfake attacks or other instances in which multiple deepfakes come from the same source.
How It Works
Before we could use deep learning to generate images, criminals and other ill-intentioned actors had a limited amount of options. Cameras only had so many tools at their disposal, and most researchers could easily identify certain makes and models. But deep learning has ushered in an age of endless options, and as a result, it’s grown increasingly difficult to identify deepfakes.
To counteract this, Facebook ran deepfakes through a fingerprint estimation network (FEN) to estimate some of their details. Fingerprints are essentially patterns left on an image due to manufacturing imperfections, and they help identify where the image came from. By evaluating the fingerprint magnitude, repetition frequency, and symmetrical frequency, Facebook then applied those constraints to predict the model’s hyperparameters.
What are hyperparameters? If you imagine a generative model as a car, hyperparameters are similar to the engine components: certain properties that distinguish your fancy automobile from others on the market. ‘Our reverse engineering technique is somewhat like recognising [the engine] components of a car based on how it sounds’, Facebook explained, ‘even if this is a new car we’ve never heard of before’.
What Did They Find?
‘On standard benchmarks, we get state-of-the-art results’, said Facebook research lead Tal Hassner. Facebook added that the fingerprint estimation network (FEN) method can be used for not only model parsing, but detection and image attribution. While this research is the first of its kind, making it difficult to assess the results, the future looks promising.
Facebook’s AI will introduce model parsing for real-world applications, increasing our understanding of deepfake detection. As cybersecurity attacks proliferate, and generative AI falls into the hands of those who would do us harm, this method could help the ‘good guys’ stay one step ahead. As Hassner explained: ‘This is a cat-and-mouse game, and it continues to be a cat-and-mouse game’.