You can’t expect Toyota to make you a safe driver
- 3-Minute Read
- For businesses relying on Microsoft, Google or other big suppliers for their AI
One of the great strengths of modern generative AI is that it’s easy to use. You just talk to it a bit like a human and the response you get back sounds much like a human too. But that is a mirage. AI is pattern-matching, based on a huge library of things that people have said or written down over the years. The output is similar, but the way of getting there is completely different.
Where your supplier can help
Toyota make excellent cars. And they have many, many features that are designed for road safety. They come with seatbelts, they can sense road conditions and some can even semi-drive for you. But they won’t stop you driving into a fast-flowing river, or into a deep pothole at night. There is only so much the car can do.
Similarly, many good suppliers take steps to ensure your AI is as safe as it can be. Microsoft, Google and others can provide mechanisms to ensure you know where your data is going, whether it will be used for training, and to make it more robust to cyber attacks.
But your supplier is not driving the car
Even with the best AI solution providers, AI Governance is essential:
- AI makes mistakes and even confidently makes things up. Staff need to know how to spot those mistakes to prevent critical business decisions being made on false information.
- AI can systematically misclassify – for example, unfairly screening out job applicants and opening up the business to a claim of discrimination.
- AI can become a replacement for thinking. Staff can produce output that looks comprehensive but contains fundamental logical flaws.
- Staff can bypass your approved AI and, for example, use free ChatGPT on their phone.
So, while it’s true that some suppliers can offset some of the risks (particularly around data residency and training), this is no replacement for appropriate AI Governance in the business. And the responsibility stays with you: under the Data Protection Act, you remain accountable for the personal data you hold, whichever supplier you use.
“We use Copilot”
There are many excellent tools on the market, such as Microsoft Copilot. But in the same way that a spreadsheet does not make you an accountant and PowerPoint does not make you a marketer, having access to Copilot does not ensure responsible AI use. You still need AI Governance. It is just that some of the risks have been partially mitigated by the supplier.
There is also a Copilot-specific risk worth knowing about. Copilot can see everything the user can see. If your file permissions are loose – and in most businesses they are – a simple question such as “what are my colleagues paid?” may surface documents nobody intended to share. Tidy up access before you switch it on.
- Key Takeaways
- Suppliers such as Microsoft can reduce some risks, but they can’t govern how your people use AI
- You remain accountable for your data, whichever supplier you use
- If you use Copilot, check who can see what before you switch it on
To explore further
Take a look at our Resources page with many other helpful articles on AI and governance
Try our tools and solutions
Discover Akili’s tools and solutions designed to help you understand,
assess and govern AI with confidence.