Explaining black-box models, Lunar timezone & Avocado Toast!
How regulation is catching up to AI models, why moon needs a timezone & a funny twist of regular avocado toast!
AI is really booming, with increased investments in large language models, image generators, automated content creation and the myriad of applications being pumped out by every company for every possible task!
These models are incredibly large, consisting of multiple layers of interconnected “neurons” which decide to fire depending on what input you’re feeding in. Not only are these models large, they are incredibly complex and rely heavily on the ginormous datasets on which they are trained upon.
As a consequence they have a black-box like reputation as the multi layered architecture coupled with colossal datasets make it unpredictable and hence, its outputs can not be explained.
With great power, comes great responsibilities. And with great responsibilities comes even greater regulation.
Regulatory bodies are trying to catch up with the advances in this technology. However, more often than not, they are sometimes heavily understaffed or really out of their domain of expertise when forming rules/regulations that affect deployment of these models.
Nevertheless, I want to point of two interesting cases wherein the regulation has caught up and is doing their best to keep it fair both for the businesses as well as the end users!
TL;DR being
being ethical in your claims
using fair datasets
FTC recently wrote a very straightforward blog, giving general good advice to anyone making claims of using AI in their products. The blog aptly titled “Keep your AI Claims in check”,asks simple questions :
Are you exaggerating what your AI product can do?
Are you promising that your AI product does something better than a non-AI product?
Are you aware of the risks?
Does the product actually use AI at all?
Answering these simple yet profound questions about product before putting up breathless marketing pitches overloaded with ostentatious words like ostentatious can really help keeping your claims in check!
Quite refreshing to study computer ethics 101 with a govt. body!
The other example is from EASA (European Union Aviation Safety Agency) ie. FAA’s European cousin which regulates various aspects of aviation.
As a part of Innovation Partnership challenge, they worked with an external company to come up with Concepts of Design Assurance for Neural Networks (CoDANN) (It even comes with a snappy abbreviation!)
This reports explores how to report on trustworthiness of a machine-learning model in a safety-critical system & how to integrate it with existing set of processes that validate the system’s operability.
The report focuses on learning assurance of a model, AI explainability and safety risk mitigation. In doing so, it proposes a W-shaped development model, which essentially an improvement over the V model typically used in automotive/aerospace development processes.
The peaks in the center focus on learning process validation, and introduces significant scrutiny of the datasets used to train the model, and how they have to also be independently verified to increase trustability.
This report finally resulted in publication of guidance for Level 1 machine learning applications.
This is indeed a great step in accepting complex machine learning models in fields where they might impact significantly, if designed with correct guidance. It is wonderful to see a regulatory body setting up advanced studies to provide useful feedback to the community which can further be translated to design guidance!
In other fun news, ESA (European Space Agency) wants to create a new time zone for the moon, to synchronize various international efforts to send people back to the moon in this decade.
Currently, moon time is based on the time of the country that is operating the certain mission. However, as we know time is always messy, especially with it running at a slightly different rate on the moon as compared to earth (56 microseconds faster, if you cared) (No, this is not a reason to have moon-light savings, if you were wondering!)
However, it looks like we are soon going to have planetary time zones, and I can already sense that a century from now, someone is going to be the complaining about them the way I do about timezones & day-light savings! 🤷
Unsolicited Recommendations:
Pod : The Soul of Music: Rhiannon Giddens excavates the past - explores origins of banjo & using history as an inspiration for writing an opera!
Book : Superfreakonomics - Unconventional wisdom presented by economist Levitt and journalist Dubner - quite an entertaining read!
Road Not Taken : Mulholland Dr - Recent LA visit gave me a chance to drive through the scenic hills surrounded by $$$ homes and some beautiful city views!
Taste Buds : Avocado Toast from Grand Central Market,RamenHood, LA
As always, thanks for reading & subscribe for more such treats !!!