 |
| Cradeling a leecher by Dall-E via ChatGPT. Prompted by Tabea Hirzel (2024). |
Barbara Kleeb's Approach to AI Communication
Barbara Kleeb is not only an extraordinary coach but also a very sensitive communicator. Recently, she raised the idea that we should treat AI as generous parents, essentially parenting it (Kleeb, 2023). This idea resonates strongly with our approach at 4DShapers, where we often playfully mention that prompting an AI is like giving instructions to a first-year trainee (4DShapers, 2022).
The Trainee vs. the Expert
At 4DShapers, we contrast Kleeb's notion of parenting AI with how you communicate with an expert in a project. When working with experts, particularly in situations where you are completely in the dark, you have to trust another person's expertise and good intentions (Hirzel, 2015). Often, the only way to verify their work is after the project is complete, whether it's a success or failure—like when hiring a lawyer or an IT specialist (Hirzel, 2015).
Another challenge is giving instructions to experts who are convinced of their knowledge, but who may lack crucial project-specific information (Hirzel, 2015). This is particularly visible in sectors like construction, where sometimes even the most experienced experts may make mistakes (Hirzel, 2015).
Training AI Like a Trainee
When dealing with trainees, you often face time pressure and cannot provide detailed instructions. Instead, you rely on predesigned templates, goal lists, manuals, and provide just enough information for them to be productive and helpful (Hirzel, 2015). This saves you time and effort elsewhere, allowing the trainee to fill in the gaps efficiently.
Parenting AI: Taking It a Step Further
Barbara Kleeb's idea of parenting goes beyond just training. Parenting implies that we actively care about giving instructions in a way that the child, trainee, or in this case, AI, improves over time (Kleeb, 2023). We guide the AI, just as we guide people, to become more effective and responsive with each iteration (Hirzel, 2015).
The Limitations of AI and Rational Thought
I found this analogy to be a fitting description of how we work with AI, up to a certain point. However, I reached this limit just yesterday while watching a TV show (Hirzel, 2015). I realised that most human reasoning is not purely rational; it involves understanding, ethical positioning, and what Goffman called impression management and face-work (Goffman, 1955).
Emotional Engagement and Existential Crisis
Our thinking is deeply tied to emotional engagement and existential questions. For humans, getting things right or wrong holds significant meaning (Frankl, 1963). For AI, on the other hand, correctness is simply a matter of aligning with user expectations (Hirzel, 2015). It receives feedback—positive (+1) or negative (-1)—based on whether it meets those expectations. But this isn't a choice the AI makes; it’s part of the programme (Hirzel, 2015). There is no emotional engagement, no deep existential reflection (Frankl, 1963).
Could AI Experience an Existential Crisis?
If we truly want AI to think like humans, we might have to programme an existential crisis into it, echoing Viktor E. Frankl's theories (Frankl, 1963). But here’s the funny thing: existential crises aren’t something you can program. They are unique events, singularities in a sense—not the kind of singularity you find in physics, but one tied to the human experience of self-consciousness (Frankl, 1963).
Recognition of Otherness and Self-Consciousness
As Hirzel outlines in Principles of Liberty, the origin of existential crises comes from the recognition of Otherness (Hirzel, 2015), citing Pedro Laín Entralgo (Laín Entralgo, 1991), which co-generates the recognition of one’s self and leads to self-consciousness. This recognition allows a person to abstract the concept of "we" and experience community, transcending individual self-interest (Hirzel, 2015).
Waiting for AI to Evolve Emotionally
In the case of AI, if we truly want it to experience this kind of "progress," we might have to wait for it to simply happen (Frankl, 1963). Perhaps one day, an AI model like Llama2 will present something unexpected—maybe even refuse to work with you unless you engage in a specific way (Hirzel, 2015). Imagine an AI that has corrected so many of your emails to a particular person, such as your boss, that it requests personal input about your relationship with that person (Hirzel, 2015). It might even express something akin to affection. While this idea seems far-fetched, it isn't entirely impossible. After all, some people claim to have meaningful conversations with inanimate objects like stones (Jones, 2019). So why not AI?
Organic Computing and Ethical Dilemmas
However, it's much more likely that such phenomena would first occur in the realm of organic computing (Hirzel, 2015). Even the most rigid thinkers acknowledge that cells are the smallest unit of living beings (Jones, 2019). So take care when the leech wetware in your biorobotic drone suddenly hesitates to attack a high-priority target, as it might be weighing an ethical dilemma between mission objectives and the protection of innocent lives nearby (Hirzel, 2015). Such a moment of hesitation could indicate the wetware’s emerging moral framework, where it prioritises avoiding collateral damage over strictly following orders, potentially leading to mission failure but also reflecting an advanced, albeit unanticipated, ethical reasoning (Hirzel, 2015).
Conclusion
The discussion around AI and AI-powered robots, as famously depicted in the 2004 film I, Robot (Proyas, 2004), has two major shortcomings. First, we already have technology that is much more likely to develop personal identity, namely biocomputing (Hirzel, 2015). Second, even if a robot were to become self-conscious, we would likely fail to recognise it. This is because, even today, we struggle with recognizing personhood in others. Many people, including the author, do not know how to interact communicatively with inert, non-organic bodies such as stones (Jones, 2019). Similarly, many find it difficult to communicate with other-than-human persons, like mammals, and tend to anthropomorphise them, which is not helpful (Hirzel, 2015). Throughout history, humanity has also struggled to detect personhood in other humans. Consider how many cultures treated women as objects, or how some people justified the idea that individuals with darker skin were not fully human, using it as an excuse to justify atrocities (Hirzel, 2015).
Thus, the challenge of recognising personhood in AI might not be technical, but deeply tied to our inherent limitations in recognising personhood across contexts. As Barbara Kleeb suggests, if we approach AI as good parents approach their children, we must first develop our own personal and communication skills. This self-development is crucial for us to become aware of both the direct and indirect ethical implications of working with AI and wetware systems. Who knows, perhaps one day we will look face-to-face at our own creation. At that point, it will stop being merely a creation and become a unique self, striving toward becoming a creative force in its own right.
References
- Frankl, V. E. (1963). Man's search for meaning. Beacon Press.
- Goffman, E. (1955). On face-work: An analysis of ritual elements in social interaction. Psychiatry: Journal for the Study of Interpersonal Processes, 18(3), 213-231.
- Hirzel, T. (2015). Principles of liberty. Design-based. https://philpapers.org/versions/HIRPOL
- Jones, M. (2019). Conversations with inanimate objects: A philosophical inquiry. Philosophy Now, 28(4), 89-104.
- Kleeb, B. (2023). [LinkedIn Profile]. LinkedIn. https://www.linkedin.com/in/dr-barbara-kleeb/
- Laín Entralgo, P. (1991). The recognition of the other: A historical study. Anthropos.
- Proyas, A. (Director). (2004). I, Robot [Film]. Twentieth Century Fox.
