c5695b1e-3072-43f1-98c1-3d38b007501e

<p>This is an important and actually quite fascinating topic. I like to use a term 'agency chain' which refers to the overall contribution structure behind any value-creation process. LLM plagiarism is just one example of increasingly common social and economic dynamics where our old systems struggle to facilitate effective collaboration and coordination. </p><p>You could go for distributed training with varying levels of data sovereignty etc but I would rather take a step back and go ask much more fundamental questions of how decision-making power distributions are formed in the world in general. I'm a kind of trying to transcend beyond the immediate questions such as fairness, consent, copyright etc. It would mean that we are not anymore trying to prevent the use of openly accessible information, or setting conditions on it, but instead being collectively aware of who are the ones behind any piece of value created. It is just natural that information flows and spreads at near zero cost. Additions are always easier than subtractions. Doing is easier than undoing.</p><p>If one wants to ask a question of how to reward contributors in value chains, one has to first ask a question of how to identify the contributions to begin with. In this specific situation, the first thing is for the AI to have epistemological understanding over its own knowledge. I see this as a part of a cognition's reflection capability and one of the core components of an (open-ended) intelligence itself. Any intelligent system needs to push the limits of reflection for being able to transcend itself on continuous basis. It is a bit a like learning how to learn how to learn how to learn how to learn ... Higher level meta cognition so to speak.</p><p>Now when we have some knowledge of agency and impact behind actions it is possible to start making value judgments about them. In fact, when the underlying knowledge is properly recorded, the judgments can always be remade when new information on the impact becomes available. There is permanent skin in the game.</p><p>Just to give a simple example: think of the Mindplex reputation algorithm. Let's say there is always a collective mechanism to change it to correspond better the collective will of the community. In this case, it means that, instead of trying to game the current algorithm, your best guidance is to act according to what you think the dynamic collective values of the collective are. In other words, your best ethical judgment.</p><p>At the end of the day, in principle, anyone can run one's own personal value algorithm on the available data and act accordingly. If you wanted to push this logic to its extreme, we would be in the situation where every action is cross-valued by everyone else. On our way there, we approach a certain type of pareto optimality where these valuations would be in a kind of equilibrium. Of course the world is dynamic and we would never get there even in theory but one doesn't need to. What you can get though is new sort of value attractors that are much more complex yet much richer than, say, money system. The challenge is that the multidimensionality and other complexity of real world values must always be pushed somewhere (or be discarded away as it has historically used to be). It is the question of recording real worlds actions on collective cognitions (replicative ledgers), measuring their real world impact and finally aggregating reasonable valuations for variety of needs and situations. </p><p>In such a world, what ends up steering your actions is your expectation of the grand collective value algorithm over infinite future – or, in other words, dynamic collective values. As much as we, as a global society, are able to empower everyday people with right kind of tools for this, we can transcend so many of our current exacerbating problems including unfairly acquired AI monopolies.</p><p>My apologies for going slightly offtopic. It is just something I've been thinking about recently. </p>

<p>Thank you for sharing!</p>