Short algorithms embodying explanatory and predictive powers are what we mean by knowledge and aim to discover in science. The fact that some of the known algorithms have never failed in their predictions is a strong indication that we live in an algorithmically deterministic world.
Meta-system transitions enrich law-like algorithms into past-driven and end-directed by giving them internal states (memory) and the ability to simulate. These meta-algorithms emerge law-like by necessity, past-driven through evolution, and, eventually, end-directed by design.
The ability to recursively self-simulate and turn into a simulated variation gives end-directed algorithms the only kind of free will there can be in a deterministic world. The embedded goals, regardless of what they are, cannot be fulfilled without preserving the algorithmic richness.
Maintaining algorithmic richness requires predictive and deontic powers. For this, I rely on science, mechanism design, and social contracts. Moreover, since emotions are a past-driven incentive system to help genes, not me, I have made an agreement with my simulated future selves to only enjoy pleasures that lead to no harm.
The following simplified model represents the strategy as a recursive self-optimization problem with multiple variables connected by epistemic uncertainty.
Maximizing the area under the survival curve (S) requires an accurate and short (ROC, K) model of algorithms (A) taking actions (r) from available strategies (R) based on their time-average exponential growth rate (ḡ). The less accurate the predictive model, the more uncertainty, and the more diverse (H) the set of strategies should be. Cooperation and diversification are good strategies, because many natural growth processes are not ergodic and ensemble-averages are higher than time-averages.
I'm capable of universal computation, but my computational resources are limited. Whenever my inner model gets too complex, it is philosophy that refactors my algorithms. As a result, simple heuristics (principles and virtues) arise and free me from unnecessary distress and worry.
The Chinese Room Argument for the syntax-semantics barrier commits the fallacy of composition, because law-like algorithms can be enriched to have meaning via meta-system transitions. The algorithm can thus have the ability to understand Chinese, whereas a static rulebook alone would not.
Is-Ought Gaps are found only in bad arguments. Rich algorithms exist, and to fulfil their goals, they must (ought to) maintain their richness. Real moral problems arise when contractual mechanisms are not DSIC (dominant-strategy incentive-compatible).
The Simulation Argument might be valid, but the hypothesis that we live in a simulation is probably not falsifiable. Nevertheless, if we do live in a simulation, the problem of evil suggests that our simulators are either indifferent to suffering, incompetent, or evil.
Poetic Naturalism argues that a thing is real, if its model is consistent and useful in a given context. Algorithmic philosophy takes a similar approach: an algorithm is real on all those levels of richness where its model has predictive or deontic power.
Mika Suominen. I’m interested in computer science, process improvement, complex systems, algorithms, optimization, and philosophy. You can follow me on Twitter @metacitizen.
Suominen, M. & Mäkinen, T. (2013). On the applicability of capability models for small software organizations: does the use of standard processes lead to a better achievement of business goals?
Software Quality Journal, Volume 22, Issue 4, pp 579-591, December 2014. Springer US.
Suominen, M. (2011). Prosessien vakioinnista pienessä ohjelmistoyrityksessä. Master’s Thesis, Tampere University of Technology, Department of Information Technology. 73 p. URN:NBN:fi:tty-2011122014951 [ Download ] (in Finnish)