Contributors   |   Messages   |   Polls   |   Resources   |  
Comments
Newest First | Oldest First | Threaded View
<<   <   Page 2 / 6   >   >>
mhhf1ve
mhhf1ve
2/9/2017 12:49:39 AM
User Rank
Platinum
Re: Because distributions matter
Wow. I can see how principles and rules like that make sense for systems that interact with physical bits -- e.g. Machines with moving parts like cars and robotic arms, etc. But I wonder if the same principles should be adopted for algorithms that just move digital bits around -- like auto trading or search algorithms or recommendation systems.

50%
50%
Ariella
Ariella
2/8/2017 8:51:40 PM
User Rank
Author
Re: Because distributions matter
@mhhf1ve precisely! that's the subject of a blog I wrote recently (awaiting publication). The concern about just this has given rise to a number of organizations looking into the issues. The Association for Computing Machinery US Public Policy Council (USACM), recently advanced 7  Principles for Algorithmic Transparency and Accountability :

1. Awareness: Owners, designers, builders, users, and other stakeholders of analytic systems should be aware of the possible biases involved in their design, implementation, and use and the potential harm that biases can cause to individuals and society.

2. Access and redress: Regulators should encourage the adoption of mechanisms that enable questioning and redress for individuals and groups that are adversely affected by algorithmically informed decisions.

 3. Accountability: Institutions should be held responsible for decisions made by the algorithms that they use, even if it is not feasible to explain in detail how the algorithms produce their results.

 4. Explanation: Systems and institutions that use algorithmic decision-making are encouraged to produce explanations regarding both the procedures followed by the algorithm and the specific decisions that are made. This is particularly important in public policy contexts.

5. Data Provenance: A description of the way in which the training data was collected should be maintained by the builders of the algorithms, accompanied by an exploration of the potential biases induced by the human or algorithmic data-gathering process. Public scrutiny of the data provides maximum opportunity for corrections. However, concerns over privacy, protecting trade secrets, or revelation of analytics that might allow malicious actors to game the system can justify restricting access to qualified and authorized individuals.

6. Auditability: Models, algorithms, data, and decisions should be recorded so that they can be audited in cases where harm is suspected.

 7. Validation and Testing: Institutions should use rigorous methods to validate their models and document those methods and results. In particular, they should routinely perform tests to assess and determine whether the model generates discriminatory harm. Institutions are encouraged to make the results of such tests public.

 

Association for Computing Machinery US Public Policy Council (USACM), though the terms differ slightly. Their Principles for Algorithmic Transparency and Accountability  are:

1. Awareness: Owners, designers, builders, users, and other stakeholders of analytic systems should be aware of the possible biases involved in their design, implementation, and use and the potential harm that biases can cause to individuals and society.

2. Access and redress: Regulators should encourage the adoption of mechanisms that enable questioning and redress for individuals and groups that are adversely affected by algorithmically informed decisions.

 3. Accountability: Institutions should be held responsible for decisions made by the algorithms that they use, even if it is not feasible to explain in detail how the algorithms produce their results.

 4. Explanation: Systems and institutions that use algorithmic decision-making are encouraged to produce explanations regarding both the procedures followed by the algorithm and the specific decisions that are made. This is particularly important in public policy contexts.

5. Data Provenance: A description of the way in which the training data was collected should be maintained by the builders of the algorithms, accompanied by an exploration of the potential biases induced by the human or algorithmic data-gathering process. Public scrutiny of the data provides maximum opportunity for corrections. However, concerns over privacy, protecting trade secrets, or revelation of analytics that might allow malicious actors to game the system can justify restricting access to qualified and authorized individuals.

6. Auditability: Models, algorithms, data, and decisions should be recorded so that they can be audited in cases where harm is suspected.

 7. Validation and Testing: Institutions should use rigorous methods to validate their models and document those methods and results. In particular, they should routinely perform tests to assess and determine whether the model generates discriminatory harm. Institutions are encouraged to make the results of such tests public.

 

100%
0%
mhhf1ve
mhhf1ve
2/8/2017 7:03:43 PM
User Rank
Platinum
Re: Because distributions matter
@Ariella - aha. Algorithms are written by human beings, so it makes sense that we'd program machines to simply amplify our biases and desires. I don't think there's a great counter argument.. unless we're using algorithms that learn somehow completely independently of us -- and come up with inscrutable methods that we don't completely understand in the end. (Ahem, Looking at you, AlphaGo!)

So do we abide by algorithms we know have biases.. or trust algorithms that we can't even understand ourselves? 

 

50%
50%
Ariella
Ariella
2/8/2017 9:06:40 AM
User Rank
Author
Re: Because distributions matter
@mhhf1ve It's somewhat more political than mathematical. Her argument is that the algorithms people point to for justifying not hiring, not allowing parole, not lending, etc. further institutitionalize discriminatory practices in society. I didn't read the whole book myself, but I did get far enough to realize one thing about the author who claims she has to take care of all the grocery shopping and meal preparation for her family because her husband is inept at it, very likely her husband is far cleverer than she thinks.

50%
50%
Ariella
Ariella
2/8/2017 9:06:39 AM
User Rank
Author
Re: Because distributions matter
@mhhf1ve It's somewhat more political than mathematical. Her argument is that the algorithms people point to for justifying not hiring, not allowing parole, not lending, etc. further institutitionalize discriminatory practices in society. I didn't read the whole book myself, but I did get far enough to realize one thing about the author who claims she has to take care of all the grocery shopping and meal preparation for her family because her husband is inept at it, very likely her husband is far cleverer than she thinks.

100%
0%
mhhf1ve
mhhf1ve
2/8/2017 2:23:32 AM
User Rank
Platinum
Re: Because distributions matter
I'll have to check out that book now. And brush up on my math skills?

50%
50%
Ariella
Ariella
2/7/2017 3:26:55 PM
User Rank
Author
Re: Because distributions matter
LOL @mhhf1ve well, mathematical proof alone is not necesarily a sufficient standard. That's the argument that underlies the book Weapons of Math Destruction.

50%
50%
Ariella
Ariella
2/7/2017 3:26:55 PM
User Rank
Author
Re: Because distributions matter
LOL @mhhf1ve well, mathematical proof alone is not necesarily a sufficient standard. That's the argument that underlies the book Weapons of Math Destruction.

50%
50%
mhhf1ve
mhhf1ve
2/6/2017 11:16:17 AM
User Rank
Platinum
Re: Because distributions matter
If only everyone held liars to the higher standard of mathematical proofs... :P

50%
50%
JohnBarnes
JohnBarnes
2/6/2017 9:50:50 AM
User Rank
Platinum
Re: Because distributions matter
mhhf1ve, batye,

There's a Godel's Theorem for Incompleteness (i.e. any non-trivial logically consistent system with a finite number of axioms must allow some unprovable-but-true statements) which implies that in all systems there are necessarily singularities, places where the system breaks down and results are undefined or nonsensical (black holes, dividing by zero, infinite informational entropy, syllogisms with "nothing" comparisons are all examples).

I would venture a guess that the reason there's no Godel's Theorem for Lying (in any non-trivial logically consistent system of communication, it will be possible for scoundrels to deceive fools) is that it's too obvious for anyone to bother to prove.

"You can prove anything with math" ignores the important point that you can also prove anything with words, pictures, or probably snow sculptures, quilts, and cake decorations, but math is uniquely checkable and verifiable. Making them use math at least forces liars to a higher level!

50%
50%
<<   <   Page 2 / 6   >   >>


Latest Articles
Italy's 5G auction could exceed a government target of raising 2.5 billion ($2.9 billion) after attracting interest from companies outside the mobile market.
The emerging-markets operator is focusing on the humdrum business of connectivity and keeping quiet about some of its ill-fated 'digitalization' efforts.
Three UK has picked Huawei over existing radio access network suppliers Nokia and Samsung to build its 5G network.
Vendor says that it's its biggest 5G deal to date.
Verizon skates where the puck is going by waiting for standards-based 5G devices to launch its mobile service in 2019.
On-the-Air Thursdays Digital Audio
Orange has been one of the leading proponents of SDN and NFV. In this Telco Transformation radio show, Orange's John Isch provides some perspective on his company's NFV/SDN journey.
Special Huawei Video
10/16/2017
Huawei Network Transformation Seminar
The adoption of virtualization technology and cloud architectures by telecom network operators is now well underway but there is still a long way to go before the transition to an era of Network Functions Cloudification (NFC) is complete.
Video
The Small Cell Forum's CEO Sue Monahan says that small cells will be crucial for indoor 5G coverage, but challenges around business models, siting ...
People, strategy, a strong technology roadmap and new business processes are the key underpinnings of Telstra's digital transformation, COO Robyn ...
Eric Bozich, vice president of products and marketing at CenturyLink, talks about the challenges and opportunities of integrating Level 3 into ...
Epsilon's Mark Daley, director of digital strategy and business development, talks about digital transformation from a wholesale service provider ...
Bill Walker, CenturyLink's director of network architecture, shares his insights on why training isn't enough for IT employees and traditional ...
All Videos
Telco Transformation
About Us     Contact Us     Help     Register     Twitter     Facebook     RSS
Copyright © 2021 Light Reading, part of Informa Tech,
a division of Informa PLC. All rights reserved. Privacy Policy | Cookie Policy | Terms of Use
in partnership with