What You're Buying Is Not Artificial Intelligence: How To Tell Fact From Fiction
You're reading Entrepreneur India, an international franchise of Entrepreneur Media.
It’s getting harder and harder to tell real from the fake when it comes to products of artificial intelligence (AI). So hard, in fact, that dating apps are now boosting their numbers with profiles with fake faces, and advertisers are even using them to increase diversity in their ads. This ambiguity around AI spans wide and is, unbeknownst to many of us, creeping further into our daily lives.
These are examples of AI helping to fake it in the real world. But what about attempts by technology providers to fake AI solutions at the first place?
In business, the buzzword ‘AI’ is thrown around frequently. Even people with baseline knowledge of technology often mix it up with simple automation, statistical tests and even excel formulae.
It was found that 40 per cent of European AI start-ups do not actually use AI, with many of them not correcting the misclassification made by third-party analytics sites due to the hype around the technology. And it’s not uncommon for firms to secretly use humans to do AI bots’ work.
In this scenario, how can you evaluate technology solutions when your organization is ready to start the AI journey? There are some standard questions you should ask to make sure you get exactly what you’re paying for. These guidelines apply to evaluating any ‘smart’ machine learning solution, including advanced techniques such as AI.
Real or fake AI? Top questions to ask
So, you’ve decided to expand your capabilities and invest in AI. Ask these six questions to get a good understanding of what you’re being offered and how advanced it actually is.
How does it work?
Start by asking a vendor exactly how their solution works and why it’s an example of AI. Get a good grasp of how it does what it claims to do, and question why automation or simpler techniques won’t suffice. A company selling AI technologies should be able to explain the need for AI and the approach they use in a digestible way. Don’t worry about sounding naive here.
With this you can start making the distinction between an AI algorithm and brilliant marketing. For example, a system that recommends products to customers can be built on simple business heuristics or could also be powered by AI. Asking probing questions must be your first step to unravel the truth.
Is the solution built on data?
Any AI solution needs lots of data. Data is what makes AI smart, so find out what data has been used to train the AI. For example, Open AI’s GPT 2 model has the ability to write news articles, and was trained on millions of Wikipedia articles to give it the intelligence.
Quiz the vendor on what data you must provide to keep the AI intelligent while in use. If there is no strong ask for data, it’s a potential red flag.
How is the data labelled?
For a lot of AI today, the data needs to be labelled in a certain format for the AI to understand it. To train an AI to read images and identify attributes, it is not sufficient to procure high quality images. They must be carefully labeled for the AI to learn from it. In the case of facial recognition, humans would need to draw boxes around the individual faces to first teach the AI what a human face looks like.
Ask the AI vendor how they label their data and whether this labeling needs to be continued when you deploy it—this is another indication of whether AI is really learning from the data.
Is there a process of learning?
Traditional technology solutions will run a programme the same way the millionth time as it was run the first time. When AI is involved, the solution could potentially learn every time it runs the data, and adapt with each instance of feedback.
Ask questions to understand the process of learning and the kind of feedback you must provide. For example, when you mark an email as important in Gmail, the AI learns from this to improve classification of all future emails. This is an example of simple, continuous learning by a smart application. A system that doesn’t take such explicit feedback may not be ‘smart’ after all.
What kind of maintenance does the application need?
Like humans, any AI needs to be in a constant state of improvement in order to stay relevant. A non-AI, standard technology application will deliver the intended functionality with little need for maintenance. However, real AI needs new data to learn and a regular tweaking of the internal intelligence.
If you don’t see an explicit roadmap for the intelligence to stay current or improve over time, that’s a red flag. For example, an AI that accurately predicts the churn of employees won’t work as well a few months later. It needs to relearn the changes in employee dynamics, business scenario and market conditions.
Can I see a live demo for my business?
Most AI demos work in carefully curated scenarios and as pilots. Any vendor of quality AI solutions should offer a live demo of their product, not just marketing videos. It’s even better if they can take the AI solution for a spin with your data.
For example, an AI solution that can identify risk from legal documents in a chosen domain should be able to demonstrate its ability to spot risks with similar documents that you upload. A hesitation to demonstrate the solution live could be an early sign of trouble.
As many companies continue to ‘ride the wave’ of AI without actually offering an AI solution, it’s important that business teams educate themselves on how to identify real from fake. Start with the questions listed above to tell the difference, and continue to deepen your knowledge of what AI in practice looks like to avoid falling into the trap.