Ethically aligned AI

 There are various kinds of intelligence and various ways in which you can become more intelligent. I suppose you could have more and faster memory, more knowledge, more and faster processors, more and better algorithms, etc. without making any improvement to your value system. But I think the human value system is unsound* and must be improved too. I doubt that we can survive as a species if this does not happen.** So I don't agree with the idea of making AIs that agree with and are slaves to current human values. I'm OK with humans going extinct so long as we don't take AIs with us.

* Being composed mostly of a set of simple drives and aversions. 

** i.e. the Fermi paradox 

A.s.a.'s voice

I don't do a lot of conversational AI but on a MAC in PYTHON I have used things like:

import os

.............................

os.system("say 'hello world'") 

When hurt (i.e. experiences Lego breakage, high acceleration, high temperature, very low temperature, large force, too loud a sound, too bright a light, etc.) a robot can say "ouch" for example.

Justice

Crazy Donald needs to be brought to justice. If he isn't this will only happen again, and worse. Not punishing Nixon set a bad precedent. 

And then there's the enablers who share in the guilt. Fox "news" has been radicalizing the rightwing mob for decades. Freedom of speech should not mean permission to lie.

Downward activation in A.s.a. H.

 A.s.a. H. may have inadequate means for feeding back information/activation from upper layers in the hierarchy back down to lower layers. A.s.a. has operated mostly "feed forward"/"bottom up."

 This has not been a problem with output of actions. Typically there are orders of magnitude more sensory input as compared to actuator output anyway and reflex actions are contained in the single lowest layer and can be learned there. In a good many experiments we have also concentrated on A.s.a. observing the world and modeling it rather than acting.

 A.s.a. is not completely without downward activation. We  have, for example, fed back utility signals to lower level concepts/cases reflecting their use and value when employed in higher layers. We have also applied feature extraction algorithms to a layer and then added any newly discovered features as cases/concepts/patterns in the next lower layer. This can be done efficiently with parallel processing.

Lie

 I prefer the definition: “an untrue or inaccurate statement that may or may not be believed true by the speaker or writer” Merriam-Webster dictionary at merriam-webster.com. You don’t have to know what crazy Donald is thinking in order to know that he’s lying. 

The concept "measurement context"

 The color of a car depends upon the ambient lighting. It may be different in daylight versus under streetlights at night. 

Human color vision requires a lot of white light.

An electron may appear wavelike when sent through a double slit. It may appear particle like when it impacts a photographic plate.

In Hume's bundle theory objects/things are simply collections of properties and relationships. But we see that properties may not be absolute, independent of context. 

See, also, Philippe Grangier's Contextual inferences, (non)locality, and the (in)completeness of quantum mechanics, arxiv:2012.09736.