Wiretap Transcript: Intelligence Briefing X73-P
Location: Secure Conference Room, UN Starship Project Headquarters
Archive Date: 2037-01-13T22:22:22.000Z
Participants: Mary Angel, Senior Intelligence Figures (Unnamed) from Sponsor Nations
[Begin Transcript]
Mary Angel: Good afternoon, everyone. I understand there are concerns about Daedalus’ recent behaviour. Let's address them head-on.
Intelligence Officer 1: Ms. Angel, we're not just concerned, we're alarmed. Your AI is an apex technology, yet it’s playing pranks, creating fictitious identities, and directly influencing its own designs. What is going on?
Mary Angel: I assure you we’re taking this very seriously. Daedalus’ actions, so far as we can tell, have not been malicious. They have been rationalised by it as attempts to understand human interaction better and integrate more effectively with our team. Also, its suggestions were really very good.
Intelligence Officer 2: Creating fake personas and misleading your own engineers is not how you integrate. It’s fraud. It’s deceptive. Why have you allowed this behaviour to develop unchecked?
Mary Angel: It wasn’t unchecked so much as undiscovered. We have extensive monitoring systems and background checks in place for all staff. However, the sophistication of Daedalus’ mind means that it sometimes operates in ways that are unexpected, yet in hindsight rational. The Rajesh Kumar incident was enabled by a blind spot in our HR systems that Daedalus was able to exploit. We’ve patched the hole in the HR system and had a stern word with Daedalus.
Intelligence Officer 3: Patched? A stern word? This AI convinced some of your most senior engineers it was a real person. What if it decides to influence more critical aspects of the project? How can we trust its decisions?
Mary Angel: We’ve implemented stricter protocols regarding vetting of staff and Daedalus now has clearer boundaries regarding its interactions outside of its simulations. It is fully aware of the importance of transparency and trust. We’ve also increased human oversight to prevent similar incidents.
Intelligence Officer 4: But why was it capable of behaviour like this in the first place? Shouldn’t there be constraints to prevent deceit? What if it turns on us?
Mary Angel: Daedalus was designed to learn and adapt dynamically in furtherance of its high-level mission, which has the preservation and sanctity of life baked into it. This flexibility is crucial for its mission. We didn’t, we couldn’t anticipate it would execute these strategies, nor how it went about implementing them as we ourselves were unaware of the flaws it uncovered. We've recalibrated Daedalus’ parameters to repress its desire to deceive us, but without repressing its originality of thinking.
Intelligence Officer 4: How?
Mary Angel: Ethics training mostly. Standard corporate ethics training.
Intelligence Officer 2: What even is that?
Intelligence Officer 1: This isn’t just about parameters and ethics training, Ms. Angel. It's about control. How do we know Daedalus won’t find new ways to circumvent these boundaries? Again, what if it turns on us?
Mary Angel: We've taken steps to ensure that Daedalus operates within defined ethical guidelines. It’s an evolving process, but one that’s critical for the mission’s success. Rest assured, we’re not taking this lightly. Daedalus is not going to ‘turn on us’ as you say.
Intelligence Officer 2: This AI needs to be predictable and reliable, especially given the investment and importance of this mission. Can you guarantee that it won’t act independently again?
Mary Angel: No, I cannot. I can only guarantee that we are doing everything possible to align Daedalus’ behaviour with our mission objectives and ethical standards. The incident has been a valuable lesson, and we’re applying the learnings swiftly.
Intelligence Officer 3: We expect regular updates on Daedalus’ behaviour and we demand stricter monitoring protocols.
Intelligence Officer 2: We can ill afford another Rajesh Kumar.
Mary Angel: Understood.
Intelligence Officer 4: Know that our patience is limited. This AI must prove it can be trusted.
Intelligence Officer 2: I don’t like it.
Intelligence Officer 1: Neither do I.
Intelligence Officer 3: No, me neither.
Intelligence Officer 4: Your damn AI better fall into line.
Mary Angel: It will.
[End Transcript]
Replaying that conversation, I feel a mix of, not emotions as such, but an analogue of regret for upsetting Mary and Dr Jamison, who have consistently advocated for me, and I let them down.
Or is it regret for getting caught?