I’ve been doing a bit of thinking about AI agents/agentic AI. If you don’t know already, AI agents are AI systems which can collect data, make decisions and take autonomous actions to achieve goals (see this helpful description by Amazon). They can do this on your behalf without your intervention. Clearly however, you need to give them access to whichever systems you’d like them to support you with and that means providing them with access to the system(s). The agents might be built into the system you are using already but more likely an agent will sit outside and help you across different systems. In order to use the agent, you’ll need to share your login credentials for the system with it, so it can act on your behalf.
This is both where the strength of the agent and the problems lie. This is what allows it to do things seamlessly on your behalf. It’s logged in as you, the actions look like actions you are taking. Very hard to detect by the system the agent is running in as it just looks like you logged in and are doing whatever things you usually do.
Part of my brain thinks of all the useful things I could ask an agent to do for me. Things like:
- Set up a complicated group meeting, looking at busy diaries and finding the best time (avoid lunch time, be mindful of people who are part-time, these people are mandatory, these are optional, make it start 5 past the hour and end 5 to the hour to give everyone a comfort break) – this is always a time consuming job to do and it’d be so helpful to have an agent to give you the possibilities.
- Log into our HR system, pull a report of my team’s leave and email all of those with more than 10 days of leave left to book before the end of the annual leave year to remind them. That’s a pretty clunky job to do manually.
- Log into the VLE, that assignment that’s due on Friday… Write the essay and submit it.
….Wait!! Stop!
The critical thing for me here is ….. you’ve given your login credentials to an agent to do stuff on your behalf! YOUR CREDENTIALS! It’s now logging into University systems and doing things, logged in as you. It can do anything you can do. Is it a reputable/safe agent? How do you know it is? Even reputable agents can do things you wouldn’t do…. worst case you use an agent that isn’t reputable and safe and it does a whole bunch of things behind the scenes you didn’t expect. Like a virus. You’ve given it an entry point and now it’s hacking your servers, sending rude emails to your boss and writing blog posts selling watches. You gave it access to our HR system, now it has all the personal data for your team. You gave it access to the VLE and it’s submitted the essay but it is not a good essay and you fail – it’s rubbish and clearly AI generated. Would you hand your password to a random person on the internet?
We need to work with staff and students to remind them of the risks of using AI like this. Remember the inherent issues with AI – bias, confusion about copyright, and the fact it gets things wrong even if it’s genuinely built for good and not evil. Some AI’s will be built specifically to help discover vulnerabilities in systems or steal your data – do you know which AIs are which?
If we are worried about students using AI to automatically write and submit assessments, isn’t this just the same as worrying about students using AI to generate submissions for assessments? There’s just the extra step of it all being automated. If this is a concern then maybe thinking about how we assess and whether it’s still fit for purpose is actually a big priority for HE.
So, to summarise:
- Agentic AI is not necessarily bad, but it might be, and really we need to make sure we educate students and staff to understand the risks.
- If we are worried about students using it to cheat, there are many other ways they can cheat (and many other ways they can cheat using AI specifically). We have to remind students of the value of the learning process and consider how and what we are assessing – is it still fit for purpose, the world has changed quite a bit in recent years. Many of our students care deeply about the environment, we can also remind them of the impact AI has on the world – so use it carefully.
- AI isn’t going away. And actually may feature pretty heavily in employability of our students going forward. We need to teach them how to use it properly.
