[isthatart]

My guess is that because the accent was put on serial computations. I may be wrong but for for me the big selling point of a(n ideal) lisp machine is that if made right you could cut the machine in half and get two lisp machines. Conversely, two lisp machines working together are a bigger lisp machine, while two Turing machines working together are not a Turing machine. Maybe now is the time for these machines, or better machine parts randomly assembled by the big blind watchmaker [0] over the web.

[0] https://en.wikipedia.org/wiki/The_Blind_Watchmaker

[chriswarbo]

> two Turing machines working together are not a Turing machine

They can be. For example multi-tape turing machines can be nice for modelling certain problems (e.g. "monotone turing machines", with a read-only tape, a write-only tape and a normal read/write tape appear a lot in algorithmic information theory).

> if made right you could cut the machine in half and get two lisp machines

I get where you're coming from in theory, but that's quite far removed from the common usage of the term "lisp machines" (i.e. those which were sold in the 80s), which weren't particularly different from other computers of the time, except for being optimised (in tandem with the OS) for particular workloads (i.e. Lisp programs).

[isthatart]

True for both comments. In the case of the "ideal" lisp machine, in my mind should be something which rewrites local patterns in memory, randomly, in many places. It is possible, theoretically. Multiheaded TMs can also be seen in the same way, provided the state of the heads are also actually on the tape. Both models are particular rewrite systems and among them the lambda calculus based one is simpler (has fewer rewrites) than the TM one. Now that we really feel the need for decentralized computing, maybe some simple generic hardware (much simpler than a TM style processor) could be more fit than what we have now.