Beyond the Hype: Senior Officers’ Toolkit for Evaluating AI Solutions

Everyone seems to be on the “AI” bandwagon these days. Indeed, if your product doesn’t have some sort of AI feature, then it may already be seen as out of date or at risk of being superseded by competitors using "AI"…

In truth, the use of AI (in all its forms) or machine learning is not new (chapeau, Alan Turing). There are many approaches, models and techniques available, depending on what you are trying to achieve.

So how do you go about sorting out the product facts from the marketing fantasy? Here’s our handy guide to making yourself a canny buyer:

1. Understand what type of AI is being used:

• ✅Ask what AI models are being used in the product. Is this a new feature, or something that has been part of the product for a long time.

• ✅Ask why the developers selected a particular model. There is always a cost/quality/reliability trade off, which is helpful to understand.

• 🚩Red Flag = “the AI is our IP / a black box”. Reliable AI providers can generally explain the underlying processes and algorithms used in their systems. A vague or evasive response to technical inquiries may signal a lack of genuine AI technology.

  2. Define your requirements:

• ✅ Understand what you are looking for the product to do. For example, do you need the product to work offline, or can it use an internet connection? Many generative AI solutions are dependent on cloud computing and storage.

• ✅ Understand what your data sources are for training the model, or can you rely on third party data. More on this later.

• 🚩Red Flag = “this is our AI product [insert product name], what is your problem?”. Understanding the available data is an essential step to model selection. The fewer models a supplier has, the more likely a square peg will be forced into a round hole. 

  3. Request a Demonstration:

• ✅Ask for live demonstrations rather than pre-recorded or scripted presentations. Real AI systems should be able to handle real-time interactions and adapt to different inputs.

• ✅Observe how the system responds to unexpected or varied inputs. Genuine AI should demonstrate adaptability and the ability to handle a diverse range of queries.

• 🚩Red Flag = Anything scripted is probably… scripted. 

  4. Verify Training Data and Sources:

• ✅Authentic AI systems are trained on diverse and representative datasets. Inquire about the sources of training data to ensure it covers a broad spectrum, avoiding biased or skewed information.

• ✅A credible AI provider should adhere to stringent data privacy and security standards.

• 🚩Red Flag = “Our [AI product] has never failed”. The AI should fail in training, just like the best of us! If the training methodology and history can’t be demonstrated or explained (ideally with things that didn’t go as expected) or if you have any concerns about how user data is handled, go deeper. 

  5. Evaluate Consistency and Limitations:

• ✅Any AI product should perform consistently within their defined capabilities. Inconsistencies or sudden drops in performance without a clear explanation may be indicative of fundamental issues, poor model validation or software quality control.

• ✅Legitimate AI providers are transparent about the limitations of their systems. If a provider claims their AI is perfect without acknowledging any shortcomings, it might be a sign of exaggeration.

• 🚩Red Flag = “Our [AI Product] is a flawless silver bullet”. It never is - do more digging. 

  6. Openness and Transparency:

• ✅Trustworthy AI providers often share information about the model architecture, training methodologies, and key algorithms. A lack of transparency may suggest an attempt to conceal the system's true nature.

• ✅Assess whether the provider adheres to ethical guidelines and practices. Ethical considerations, such as avoiding bias in algorithms and ensuring user privacy, are crucial indicators of a responsible AI approach.

• 🚩Red Flag = Hiding behind an Intellectual Property shield. 
  

By employing a combination of these approaches, both individuals and organisations can better assess the authenticity of AI claims and distinguish between genuine AI technology and instances of misrepresented (mistakenly or otherwise) AI capabilities.

If you need help navigating what is a fast evolving and pretty complex landscape, or want a better understanding of the value these products are (and will) bring… reach out to us: hello@missiondecisions.com