Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
About the Author
About the Author
About the Author

Ethical Enterprise AI: A Siren Call and Five Fold Path for Business, Tech and HR Leaders to Show the Way

Being Human for a Better Tomorrow in the Age of AI
  • October 4, 2024
  • 18th edition and 7th in Vol 2
Ethical Enterprise AI scaled

The previous edition of the newsletter presented a framework adapted from Sun Tzu’s Art of War for addressing challenges such as data security in the Age of Intelligence with conviction, clarity, and wisdom.

Ethical use of data and AI for inferences goes beyond security and compliance into values and principles of how data is collected, analysed, interpreted, and shared.

This is not a theoretical discussion but a siren call for what is urgent and concerning. Consider the following recent public examples where ethics took a back seat—or is yet to take a seat:

1. LinkedIn’s AI Training Practices

LinkedIn set a default yes to share user data for AI training without obtaining clear, informed consent. What is worse, users weren’t alerted about this significant change.

2. 23andMe’s Genetic Data Concerns

With 14 million Americans’ genetic data on hand, 23andMe faces scrutiny over potential buyouts, data breaches, and concerns about ownership and misuse of deeply personal information.

3. Wildchat’s Public Dataset

1 million ChatGPT transcripts, although collected with consent, contain private details about users’ health, relationships, and more.

Gen AI chat is always available when you need, is willing to listen, and doesn’t judge—and that is perhaps why people trust them with very private information and thoughts more than friends or family.

While no misuse has occurred yet, history tells us this will become a problem over time.

4. Clinical Diagnostic AI Performance

A clinical study to find effectiveness of large language models in diagnostic reasoning found GPT-4 demonstrated higher performance than doctors.

However, the ChatGPT data also included curated vignette diagnostics in the data, and that overstated ChatGPT performance.

5. Employee Monitoring using AI

A CNBC article describes how firms like Walmart, Delta, Starbucks, T-Mobile, and Chevron are using AI to monitor employee messages for a wide range of purposes like sentiment monitoring, identifying harassment, discrimination, etc.

There are ethical concerns including privacy violations, lack of transparency, impact on mental health from constant surveillance, and potential misuse.

The Five Fold Path for Ethical Enterprise AI

As is often in our explorations at Mindvista, ancient wisdom provides answers to these modern questions.

For instance, many religions and non-theistic practices suggest a five fold path for joy, fulfillment, and liberation. Likewise, there is also a five fold path to provide principles and ethical guidance to business and tech users as outlined below:

Principle #1: Get informed consent

Go beyond all-or-nothing fine print agreements, establish tradeoffs, and get explicit informed consent for the subject to share data.

In the LinkedIn example above, they should have provided a benefit to users to share data for AI training and taken explicit informed consent.

Principle #2: Ensure representative dataset

Ensure datasets represent the past and can be used to extrapolate future trends.

In the clinical study example above, the ChatGPT data was not representative of what doctors would typically see. Likewise, if AI is used to filter resumes—as an Applicant Tracking System—and past data has gender or race bias, then it may not be representative of future intention.

Principle #3: Model correction for bias and behaviour changes

Model correction must account for bias, changes in user behaviour, and judgment of inferences.

Before AI, Google Flu Trends (GFT) is a good example for the need for model correction.

GFT was launched in 2008 to forecast influenza activity by analyzing search query data related to flu symptoms. Initially, it succeeded in predicting flu outbreaks faster than traditional methods.

However, in early 2013, GFT overestimated doctor visits for influenza-like illness—predicting more than double the actual cases reported by the CDC.

This failure was due to changes in search behaviour and the model’s inability to adapt, leading to a loss of credibility.

Principle #4: Periodic review and controls

Conduct periodic review and establish controls on data use and inferences for validity and relevance.

Humans make judgment errors—and systems do too. Credit scores in the past had errors in their calculations.

In a random sample, a surprising 44% of consumers found errors in their credit reports (USA Today), despite a well-defined and regulated process.

AI systems will make errors, and periodic review and control on data use is key. In the Google Flu Trends example, Google discontinued it in 2015.

Principle #5: Build trust and consensus

Especially where technology is moving ahead of users and law.

In the case of employee monitoring, establish basis, valid and invalid uses, benefits, and common good to win trust and get consensus.

In the Wildchat example—or brain neuronal mapping—we don’t need to wait for misuse to happen. We must proactively create awareness, establish security, and ring-fence sensitive data to prevent misuse.

While these are public examples, the principles of the Five Fold Path apply for every employee- or customer-facing use case in every company.

A centralized or federated Ethics Review Board to examine, authorize, and prevent misuse would be a good institutional mechanism to implement if not already in place.

Extending beyond, regulators in the EU and India mandate ESG (Environmental, Social, and Governance) reporting by law.

Asking for checks, balances, and reporting on ethical use of AI is not a far-fetched ask and has as much consequence for social good as ESG.

Utilitarianism is a tradition of ethical philosophy founded by Jeremy Bentham and John Stuart Mill that advocates actions for the greatest amount of good for the greatest number of people.

Ethical AI is Utilitarianism in the Age of Intelligence, and leaders should show the way.

PS: In a major breakthrough, a group of scientists have just published the complete 140,000 neuron map of DragonFly.

The model simulates all observed cause and effects in real flies. This raises profound ethical questions—but in a lighter vein, if we can handle the current level of AI technology ethically, there’s hope for us when neuronal mapping becomes a reality!

Best wishes

Select Quotes

While I have some background in data security and technology, hyphenating ethics to data science was entirely new.

I learnt a lot from H. V. Jagdish, Professor of Electrical Engineering and Computer Science at the University of Michigan and Distinguished Scientist at the Michigan Institute of Data Science.

His teaching, clarity of thought, wisdom, and conviction are exceptional and helped me understand, relate, and synthesize this article.

The quotes below are from him and his teaching:

"All or nothing is not a good approach; better to provide graduated choices so users can make tradeoffs they like."
"Different boundaries do not mean no boundaries."
"Feel our way through social consensus for high volume social impact consensus."
"Compliance is about what you must do and ethics is what you should do."
"Two point Data Ethics Code:
  • Do not surprise the subjects.
  • Own the outcomes."
“AI is a language. Treat it like one: practice, iterate, and mind your grammar prompts, assumptions, and verification.”

II. Questioning / Asking

Good conversations flow from well‑sequenced questions—topical, simple, coherent, cohesive.
LLM Conversation Example 1
Q: What are empirical judgments?

A: Empirical judgments are based on observation, experience, or experimentation.
Q: What are moral judgments?

A: Moral judgments are based on ethical principles and values.
Share: