HOW CAN I TRUST AI?

/in /by

How far can we trust AI?

We often just don’t know if AI is right, wrong, hallucinating, or biased. 

OK. How can we tell when to trust, when to check?

You can’t 100 per cent tell when AI is 100 per cent right–but you can reduce your risks while still gaining the tremendous benefits of the knowledge and capabilities of ChatGPT, CoPilot, or your preferred AI app.

The Risks–The Remedies

Errors in facts:

We’ve all heard about AI making up facts, whether that be court cases, research citations, biographic data about historical or current people, even answer keys–you can probably tell give me other examples.

Yet if your submitted work has AI-generated errors, you are still accountable for the accuracy of your work. 

Remedies?

  • Check citations and other information with original documents and sources.
  • Use more than one AI app and check them against each other.
  • Open declaration about where AI has been your source and what you have done to verify the information maintains your credibility even in the case of unwise reliance. 

Errors in analysis:

Errors in evaluating alternatives or critiquing opinions, analyses, and theses are trickier to check. If the AI results are based on wrong or hallucinated information, you won’t spot that possibility at first glance. This can make you look inept in the eyes of those with more experience (like your instructors), or can lead you to a less-than-ideal financial decision.

Remedies? 

  • Probe further, asking for sources and reasons for the points in the analysis, which you can check as mentioned above.
  • Again, using more than one AI reduces your risks and gives you more to think about (which is not a bad thing, what with thinking being what students do). 
  • It also helps to know the special strengths of the different apps in the subject area you are investigating.

Pandering to bias:

ChatGPT is one AI app that has been remarked as complimenting people on the brilliance of their prompts and comments, and tailoring responses to the data they have from your previous queries, as well as whatever other data it has collected on you. If you only want echo chamber responses, okay. Otherwise, seek remedies and see “Security” below.

Remedies? This is a tough one.

  • Ask for additional analysis with prompts that counter bias: “How would someone who disagrees with that analysis make a strong case?”
  • And again, use more than one app.
  • Use additional security to make it difficult for the app to “read” your biases. 

Security:

Information about your interests, your studies, your web traffic, and much else can be available to the AIs you use. One of my colleagues, questioning an invoice from a double-billing supplier, asked AI how to respond to the nth request for double payment. She is now face-palming at the response, which identifies her as a slow-pay who needs to guard her credit rating. And that comment is now in that app’s database, and who knows where else. Ouch!

Remedies?

  • Clear your history from the app after each use.
  • Use the app without signing in to your account. Use it without creating an account if you can.
  • Use the apps through virtual internet access. This is easy to set up and is often free — but watch what the provider does in terms of privacy.
  • Watch what information you are revealing to your app through your prompts. 

Seek expertise:

AI performance capabilities and tracking of your data capabilities are both increasing exponentially. Stay up-to-date on the uses, risks and remedies by regularly monitoring information sites and newsletters with up-to-date, accessible information geared to your level of technical understanding.

Mandatory: 

  • Always, always, always stay up to date on what your campus allows, forbids, and encourages on the use of AI.
  • Keep documentation that proves your virtue. 

Check in with me [email protected] to share what you have learned from using AI, and to work together designing its creative, strategic use. The goal is to use AI to effectively gain your academic goals.

The information in this blog cannot take the place of support from your own mental health professional or community health resources. Reach out to them. And IF YOU ARE IN CRISIS, PLEASE DIAL 911.