Your point about individuals with extreme (objective) vulnerabilities is especially well-taken. There is, however, some room to mitigate oversensitivity to individual cases. As you note, in Section II.B, I suggest that the “E” parameter, which measures rough equality in pairwise comparisons, can be treated as a function, in part, of aggregate welfare loss or gain. The idea is that we are less willing to cater to individual needs when the aggregate loss from doing so is great. But the sensitivity cannot be too high, or else the whole point of disaggregated comparisons is lost. The model is meant to be flexible enough to accommodate different moral intuitions. The main goal is to capture the factors that clearly matter, rather than to fix their relative weights once and for all.

Your interpretation of my view is, on the whole, exactly right. There may, however, be a small misunderstanding behind your second comment. You note that individuals with idiosyncratic preferences might be over-privileged in the analysis. But the model is not meant to be subjective in its assessment of individual burdens. The pairwise comparisons can be conducted with as objective a measure of individual burden as one likes—for example, in terms of quantifiable costs incurred. The same goes for individual differences in values or priorities. You’re right that doing justice in light of such differences among people is difficult, but that is everyone’s problem. For instance, the model treats everyone as equally risk-averse — an extreme idealization — but one that seems unavoidable if we are to make objective assessments at all.

Your point about accumulating costs in the design of complex goods is also well-taken, especially since a single-issue lawsuit focused on one set of plaintiffs is unlikely to reflect on lawsuits down the line. One limiting principle already built into the model is that the costs and benefits of a design change must be knowable at the outset. As I note early on, the analysis concerns known impositions of risk. But you’re right that additional limiting principles or doctrines will likely be needed. Perhaps, given time constraints, defendants should only be required to deliver a careful, morally sound verdict based on disaggregated Hand on the most glaring and feasible design changes—those visible to the eye of ordinary (corporate?) vigilance.

I look forward to thinking more about these issues. I continue to find the Hand formula remarkably useful for structuring moral reflection. One of my aims in the paper was precisely to make its subtle flexibility and complexity more vivid.

Hell of complicated issues here. But, one thing right now:

The legal philosophy of Sec 230 was among others, the creation of new horizons in terms of free speech (thanks to the internet obviously). Free speech, is justified among others, by enriching the speech or free market of exchange of opinions let’s say.

So, one may argue, that as well when dealing with AI indeed. The AI technology, can create new horizons in free speech terms. The ability, to swiftly connect very remote dots and manipulate huge volume of information (manipulate in the positive sense rather) would create new horizon in that free market of exchange of opinions.

Liability ? Well, the internet has increased potential liability, thanks, or due to, anonymity. Anonymous comments, defaming endlessly other persons, and evading simply liability. But, for the sake of free speech, they are getting away with it, and stay or still reside behind the veil of unknown commenters.

The same one may argue, concerning AI.

Thanks