permacomputing

Source repository for the main permacomputing wiki site
git clone http://git.permacomputing.net/repos/permacomputing.git # read-only access
Log | Files | Refs

artificial_intelligence.mdwn (2002B)


      1 Artificial intelligence
      2 =======================
      3 
      4 **Artificial intelligence** (**AI**) refers to machine actions that are
      5 perceived by humans as intelligent. **Machine learning** is a subset of AI
      6 where the "intelligence" is not programmed in but learned from data by a
      7 machine learning model. In current mainstream usage, these terms are nearly
      8 synonymous, but AI in general is a very vast field of different approaches,
      9 most of which are quite obscure to non-experts.
     10 
     11 Dumb and smart programs
     12 -----------------------
     13 
     14 Humans instinctively relate to machines and tools either as body extensions
     15 or as autonomous creatures. It is generally not a good idea to require a
     16 user to use both approaches simultaneously. "Smartness" is not wanted in
     17 applications that are supposed to be wielded as tools or instruments, if it
     18 makes them more complex and less predictable. If there are "smart" functions
     19 in a tool, they should be clearly separate from the "wielded" portion, with
     20 the option of disabling them completely.
     21 
     22 Among the most important software, [[compiler]]s are programs that are
     23 generally supposed to be rather smart in order to produce efficient code for
     24 the target platform. It is also where a large amount of resource use can
     25 often be justified by the energy that is saved by the efficiency of the
     26 produced code.
     27 
     28 Green AI
     29 --------
     30 
     31 The research, training and deployment of very large machine learning models
     32 takes a radically increasing amount of energy and dedicated hardware in
     33 today's world. This is why "Green AI" has become a thing.
     34 
     35 The [tinyML foundation](https://www.tinyml.org/) is concerned about small
     36 machine learning models that run on very low power when trained (but may
     37 still require a lot of resources to train).
     38 
     39 (Please include interesting information/resources about low-power AI/ML, if
     40 you have studied this topic)
     41 
     42 See also:
     43 
     44 * [Schwartz&al's Green AI paper from 2019](https://arxiv.org/abs/1907.10597)
     45 * [Playing Atari with Six Neurons](https://arxiv.org/abs/1806.01363)