The Excessions of xAI : Algorithms, Ethics, Responsible AI and the Explainability — Part 1 of 7
Planning on a se7en part series on this topic, with a focus on pragmatic aspects of practicing xAI/Responsible AI
Theoretical explorations and precise definitions are necessary, but not sufficient. While precise definitions don’t help us, we can’t shun any form of definitions as without definitions we can’t conceptualize what we are doing or going to do !
Prof. Kartik Hosanagar in his insightful book A Human’s Guide To machine Intelligence has done an excellent job of articulating the nature of the beast — the unanticipated (and complex) consequences of AI, what causes the algorithms to behave in an unpredictable manner and finally when, where and how to use AI as well as shape the narrative of why it does what it does. A must read. My book review here.
Interestingly, am speaking at the Twin Cities Ethical Conference organized by The University of St.Thomas’ Center for Ethics in Practice. The topic is “The Sense and Sensibility in the Ethical use of Artificial Intelligence”. [Link] Thus this blog is my sounding board for the talk !
I will make the talk a lot more interesting … don’t judge a talk by the blog ;o)
What does a Responsible AI do for a living ? - you might ask. Naturally, it depends on whom you ask and what you expect … therein lies the rub … and that is the topic of this blog series ! Plan to elaborate each one of the following pragmas in a separate blog …
#1 : Know what you want out of AI — Define your expectations !
AI is a distinct tool, not a panacea (yet!) It can’t yet replace humans or even get close … In Part 3, We will take a look at some of the ways we can define AI without really defining it !
#2 : Know the Trajectory, Attack Vectors & the Unreasonable Effectiveness of AI
Interestingly all the hype about AI results from this pragma. AI has an exponential-exponential trajectory; unfortunately humans can’t grok exponentials at all ! That is why AI takes us by surprise and create unintended attack vectors. We mistake an exponential AI as the evidence of a sentient object … not yet … We will examine some of the interesting asymptotes of AI, in Part 4.
The ultimate cost of hiding complexity can be much more than the complexity itself !
Remember, we estimated that AI will be able to play the Go game well, only by the year 2030 ! Now it is putting Go world champions out of business !
#3 : Know the effects of Algorithms, Models, Data, Decisions and Reasoning
In the bygone years, we could precisely list the steps to do a thing, call it an algorithm and the decisions it makes are very clear to us.
Unfortunately, no more …
Now we define an algorithm as an architecture … i.e. how to string a bunch (sometimes a yuuge bunch) of neurons together and then we let them loose over a large pile of data .. they learn and the result is a model … when we give it actual data, it churns out decisions
- Model := Algorithm + Training Data
- Decision := Model + Actual Data
We don’t exactly know what they have learned, don’t have a clue how they make decisions and more problematic, we can’t reason about their decisions like we do with algorithms … Of course we have tools, yes we have practices … and yep the silicon valley venture capitalists are minting new shiny startups in this space … all of which we will explore in Part 5 (I think) …
#4 : Know the Larger Principles of Ethical use of AI & the subset you care about
Now that we have an idea what an AI is and can estimate it’s power, we can talk about Responsible AI and Ethics. But be careful about getting deep into moral discourses and objective philosophy !
You do need to resolve Non-maleficence vs. Beneficence; you need to understand what “Primum non nocere” means … and why “first, do no harm” is much better than “Do good”
Say you are a physician and has four patients who will be healthy … only if they could find donors for different organs … and … the orderly wheels in a very sick 5th patient with a very low survival chance, but can donate the badly needed 4 organs ! Would you exchange the well being of 4 patients for this single patient ? Now you realize the relevance of the Hippocratic Oath— The movie “I am mother” does an excellent job articulating this problem
#5 : Know how to operationalize — explainability and the rest of the **ilities viz. interpretability, transparency, contestability,…
Yes, we do have to come down from the elevations and make these ideas real ! And that is the topic for our Part 7 !
In short, we will dig deeper into these 5 pragmas in future blogs …
Before we jump into the gory details, let us seek Ethical AI from three very interesting authors, in the next three sub-parts (2a,2b,2c)
- First let us hear from Prof. Kartik Hosanagar via his insightful book A Human’s Guide To machine Intelligence — Done ! Here
- Then we will visit The Ethical Algorithm by Michael Kearns and Aaron Roth
- We will finish our trek by conversing with Gary Marcus and Ernest Davis via their book Rebooting AI: Building Artificial Intelligence We Can Trust — Done ! Here
Till then adios …
