AI and Ethics Archives - AiThoughts.Org

Ethics at the Forefront: Navigating the Path at frontier of Artificial General Intelligence

Posted on February 2, 2024 - 8:09 am by Saji Thoppil

While we may want to cautiously avoid the term Artificial General Intelligence (AGI) today, it is evident from the general capabilities of the systems currently in place that we are either close to, or perhaps already have, some form of AGI in operation. In this scenario, it is crucial that AI ethics take center stage for several compelling reasons:

Unprecedented Autonomy and Decision-Making: AGI’s capability to autonomously perform any intellectual task necessitates ethical guidelines to ensure that the decisions made do not harm individuals or society.

Societal Impact and Responsibility: The profound impact of AGI across all societal sectors demands an alignment with human values and ethics to responsibly navigate changes and mitigate potential disruptions.

Avoiding Bias and Ensuring Fairness: To counteract the perpetuation of biases and ensure fairness, AGI requires a robust ethical framework to identify, address, and prevent discriminatory outcomes.

Control and Safety: The potential for AGI to surpass human intelligence necessitates stringent ethical guidelines and safety measures to ensure human control and to prevent misuse or unintended behaviors.

Transparency and Accountability: Given the complexity of AGI decision-making, ethical standards must enforce transparency and accountability, enabling effective monitoring and management by human operators.

Long-term Existential Risks: Aligning AGI with human values is crucial to avert existential risks and ensure that its development and deployment do not adversely impact humanity’s future.

Global Collaboration and Regulation: The global nature of AGI development necessitates international cooperation, with ethical considerations driving dialogue and harmonized regulations for responsible AGI deployment worldwide.

To expand on the important aspect of “Unprecedented Autonomy and Decision-Making,” the profound ability of AGI systems to perform tasks across various domains without human intervention is noteworthy. Organizations can proactively establish certain measures to ensure that the development and deployment of AI systems are aligned with ethical standards and societal values. Here’s what organizations can put in place now:

Aspect	Manifestation	Importance of Ethics
Decision-Making in Complex Scenarios	Future AGI can make decisions in complex, unstructured environments such as medicine, law, and finance.	Ensuring Beneficence: Ethical guidelines are needed to ensure decisions made by AGI prioritize human well-being and do not cause harm.
Continuous Learning and Adaptation	Unlike narrow AI, AGI can learn from new data and adapt its behavior, leading to evolving decision-making patterns.	Maintaining Predictability : Ethical frameworks can guide the development of AGI to ensure its behavior remains predictable and aligned with human intentions.
Autonomy in Execution	AGI systems can act without human oversight, executing tasks based on their programming and learned experiences.	Safeguarding Control: Ethics ensure that even in autonomous operation, AGI systems remain under human oversight and control to prevent unintended consequences.
Interaction with Unstructured Data	AGI can interpret and act upon unstructured data (text, images, etc.), making decisions based on a wide array of inputs.	Preventing Bias: Ethical standards are crucial to ensure that AGI systems do not perpetuate or amplify biases present in the data they learn from.
Complex Communication Abilities	AGI can potentially understand and generate natural language, enabling it to communicate based on complex dialogues and texts.	Ensuring Transparency: Ethics demand that AGI communication remains transparent and understandable to humans to facilitate trust and accountability.
Long-Term Strategic Planning	AGI could plan and execute long-term strategies with far-reaching impacts, considering a wide array of variables and potential future scenarios.	Aligning with Human Values: Ethical guidelines are essential to ensure that AGI’s long-term planning and strategies are aligned with human values and ethics.

By taking these steps, organizations can play a pivotal role in steering the development of AGI towards a future where it aligns with ethical standards and societal values, ensuring its benefits are maximized while minimizing potential risks.

Insights into AI Landscape – A Preface

Posted on November 1, 2023 - 4:49 pm by Anil Sane

AI Landscape and Key Areas of Interest

The AI landscape encompasses several crucial domains, and it’s imperative for any organization aiming to participate in this transformative movement to grasp these aspects. Our objective is to offer our insights and perspective into each of these critical domains through a series of articles on this platform.

We will explore key topics each area depicted in the diagram below.

1. Standards, Framework, Assurance: We will address the upcoming International Standards and Frameworks, as well as those currently in effect. Significant efforts in this area are being undertaken by international organizations like ISO, IEEE, BSI, DIN, and others to establish order by defining these standards. This also encompasses Assurance frameworks, Ethics frameworks, and the necessary checks and balances for the development of AI solutions. It’s important to note that many of these frameworks are still in development and are being complemented by Regulations and Laws. Certain frameworks related to Cybersecurity and Privacy Regulations (e.g., GDPR) are expected to become de facto reference points. More details will be provided in the forthcoming comprehensive write-up in Series 1.

2. Legislations, Laws, Regulations: Virtually all countries have recognized the implications and impact of AI on both professional and personal behavior, prompting many to work on establishing fundamental but essential legislations to safeguard human interests. This initiative began a couple of years ago and has gained significant momentum, especially with the introduction of Generative AI tools and platforms. Europe is taking the lead in implementing legislation ahead of many other nations, and countries like the USA, Canada, China, India, and others are also actively engaged in this area. We will delve deeper into this topic in Series 2.

3. AI Platforms & Tools: AI Platforms and Tools: An array of AI platforms and tools is available, spanning various domains, including Content Creation, Software Development, Language Translation, Healthcare, Finance, Gaming, Design/Arts, and more. Generative AI tools encompass applications such as ChatGpt, Copilot, Dall-E2, Scribe, Jasper, etc. Additionally, AI chatbots like Chatgpt, Google Bard, Microsoft AI Bing, Jasper Chat, and ChatSpot, among others, are part of this landscape. This section will provide insights into key platforms and tools, including open-source options that cater to the needs of users.

4. Social Impact: AI Ethics begins at the strategic planning and design of AI systems. Various frameworks are currently under discussion due to their far-reaching societal consequences, leading to extensive debates on this subject. Furthermore, it has a significant influence on the jobs of the future, particularly in terms of regional outcomes, the types of jobs that will emerge, and those that will be enhanced or automated. The frameworks, standards, and legislations mentioned earlier strongly emphasize this dimension and are under close scrutiny. Most importantly, it is intriguing to observe the global adoption of AI solutions and whether societies worldwide embrace them or remain cautious. This section aims to shed light on this perspective.

5. Others: Use Cases and Considerations: In this Section, we will explore several use cases and success stories of AI implementation across various domains. We will also highlight obstacles in the adoption of AI, encompassing factors such as the pace of adoption, the integration of AI with existing legacy systems, and the trade-offs between new solutions and their associated costs and benefits. We have already published a recent paper on this subject, and we plan to share more insights as the series continues to unfold.

The Executive Order!

Posted on November 1, 2023 - 6:20 am by Diwakar M

Close on the heels of the formation of the Frontier Model Forum and a White House announcement that it had secured “voluntary commitments” from seven leading A.I companies to self-regulate the risks posed by artificial intelligence, President Joe Biden, yesterday issued an executive order regulating the development and ensuring safe and secure deployment of artificial intelligence models . The underlying principles of the order can be summarized in the picture.

The key aspects of the order focus on what is termed “dual-use foundation models” – models that are trained on broad data, uses self-supervision, and can be applied in a variety of contexts. Typically the generative AI models like GPT fall into this category, although, the order is aimed at the next generation of models beyond GPT-4.

Let’s look at what are the key aspects of what the order says in this part. Whilst the order talks about the

Safe & Secure AI

The need for safe and secure AI through thorough testing – even sharing test results with the government for critical systems that can impact national security, economy, public health and safety
Build guidelines to conduct AI red-teaming tests that involves assessing and managing the safety, security, and trustworthiness of AI models
The need to establish provenance of AI generated content
Ensure that compute & data are not in the hands of few colluding companies and ensuring that new businesses can thrive [This is probably the biggest “I don’t trust you” statement back to Big Tech!]

AI Education / Upskilling

Given its criticality, the need for investments in AI related education, training, R&D and protection of IP.
Support for programs to provide Americans with the skills they need for the age of AI and attract the world’s AI talent, via investments in AI-related education, training, development, research, and capacity and IP development
Encouraging AI skills import into the US [probably the one that most Indian STEM students who hope to study and work in the US will find a reason to cheer]

Protection Of Rights

Ensuring the protection of civil rights, protection against bias & discrimination, rights of consumers (users)
Lastly, also the growth of governmental capacity to regulate, govern and support for responsible AI.

Development of guidelines & standards

Building up on the Blueprint AI Bill of Rights & the AI Risk Management Framework, to create guidance and benchmarks for evaluating and auditing AI capabilities, particularly in areas where AI could cause harm, such as cybersecurity and biosecurity

Protecting US Interests

The regulations also propose that companies developing or intending to develop potential dual-use foundation models to report to the Govt on an ongoing basis their activities w.r.t training & assurance on the models and the the results of any red-team testing conducted
IaaS providers report on the security of their infrastructure and the usage of compute (large enough to train these dual use foundation models), as well as its usage by foreign actors who train large AI models which could be used for malafide purposes

Securing Critical Infrastructure

With respect to critical infrastructure, the order directs that under the Secretary Homeland Security, an AI Safety & Security Board is established, composed of AI experts from various sectors, to provide advice and recommendations to improve security, resilience, and incident response related to AI usage in critical infrastructure
All critical infrastructure is assessed for potential risks (vulnerabilities to critical failures, physical attacks, and cyberattacks) associated with the use of AI in critical infrastructure.
An assessment to be undertaken of the risks of AI misuse in developing threats in key areas like CBRN (chemical, biological, radiological and nuclear) & bio sciences

Data Privacy

One section of the document deals with mitigating privacy risks associated with AI, including an assessment and standards on the collection and use of information about individuals.
It also wants to ensure that the collection, use, and retention of data ensures that privacy and confidentiality are respected
Also calls for Congress to pass Data Privacy legislation

Federal Government Use of AI

The order encourages the use of AI, particularly generative AI, with safeguards in place and appropriate training, across federal agencies, except for national security systems.
It also calls for an interagency council to be established to coordinate AI development and use.

Finally, the key element – keeping America’s leadership in AI strong – by driving efforts to expand engagements with international allies and establish international frameworks for managing AI risks and benefits as well as driving an AI research agenda.

In subsequent posts, we will look at reactions, and what it means for Big Tech and for the Indian IT industry which is heavily tied to the US!

AI and Law

Posted on September 23, 2023 - 3:28 pm by L Ravichandran

The Public Domain is full of initiatives by many Law Universities, large law firms, and various government departments on the topic of “AI and Law “. I was happy to see a news article a few days ago about the Indian Consumer grievances cell thinking about using AI to clear a large number of pending cases. They have had some success in streamlining processes and making it all digital but they felt that the sheer large volume of pending cases needs AI-type intervention. I have already talked about the huge volume of civil cases pending in lower courts in India and some cases taking even 20 years to get final judgment. As the saying goes “Justice delayed is Justice denied”, it is imperative that we find solutions to this huge backlog problem.

All discussions are centred around two broad areas: –

1. Legal Research and development of customer’s case by Law firms. Basically, core work of both junior and senior law associates and partners.

2. Assisting judges or even rendering judgment on their own by AI models to reduce backlog and speedy justice.

Lots of interesting discussions happening on (1). Law research, looking into archives, similar judgments, precedence’s, etc. seem to be a no-brainer. Huge advances in automation have been already done and this will increase multi-fold by these Law purpose-built language models. What will happen to junior law associates is an interesting question. Can they use better research and develop actual arguments and superior case brief for their clients and take the load off senior associates who in turn can focus more on customer interactions? I found discussions on the model analysing judges’ earlier judgments and making the argument briefs customized per judge, fascinating.

The no (2) item needs lot of discussions. All democratic countries jurisprudence is based on these 3 fundamental principles.

Every citizen will have their “day in the court” to present their case to an impartial judge.
Every citizen will have a right to a competent counsel with a provision of public defenders given free to the citizens.
Every witness can be cross examined by the other party without any restrictions.

On the one hand, we have these great jurisprudence principles. On the other hand, we have huge backlogs and delays.

How much citizens are willing to give up some of the basic principles to get speedy justice?

Can we give up the principle of “my day in Court” and let only written briefs submitted to the court to be used for final judgement? This will mean witness statements in briefs will not be cross examined or questioned.

Can we give up the presence of a human judge who will read the briefs on both sides and make a judgement and let an AI Model read both the briefs and pronounce the judgement?

Even if citizens are willing to give up these principles, does the existing law of the land allow this? It may require changes to law and in some countries even changes to their constitution to allow for this new AI jurisprudence.

Do we differentiate between civil cases and criminal cases separately and find different solutions? Criminal cases will involve human liberty issues such as imprisonment and will need a whole set of different benchmarks.

What about changes to appeal process if you do not like lower court judgment? I presume we will need human judges to review the judgements given by AI Models. It is very difficult for us to accept higher court AI model, reviewing and correcting a lower court AI model’s original judgement.

The biggest hurdle is going to be us, the citizens. In any legal case involving two parties, one party always and in many cases both parties will be unhappy with any judgement. No losing party in any civil case is going to be happy that they lost as per some sub clause in some law text. In many cases, even winning parties may not be happy with the award amount. In this kind of scenario, how do you expect citizens to accept an instantaneous verdict after both parties submit their briefs? This will be a great human change management issue.

Even if we come out with some solutions to these complex legal and people problems, one technical challenge still remains a big hurdle. With the release of many large language models and APIs, many projects are happening to train these LLMs on specific domain. A few days ago, we saw a press release by EY about their domain-specific model developed with an investment of US$1.4 Billion. Bloomberg announced a BloombergGPT, their own 50-billion parameters language model purpose-built for finance. Who will bell the cat for the Law domain? Who will invest large sums of $$s and create a Legal AI Model for each country? Until this model is available for general use, many of the things we discussed will not be possible.

To conclude, there are huge opportunities to get business value out of the new AI technology in the Law and Justice Domain. However, technical, legal and people issues must be understood, addressed and resolved before any large-scale implementation.

More Later. Like to hear your thoughts.

L Ravichandran

AI Regulations : Need for urgency

Posted on August 18, 2023 - 5:57 am by L Ravichandran

Few weeks ago, I saw a news article about risks of unregulated AI. The news article quoted that in USA, Police came to a house of a 8 months pregnant African American lady and arrested her due to a facial recognition system identified her as the theft suspect in a robbery. No amount of pleading from the lady about her advanced pregnancy condition during the time of robbery and she just could not have committed the said crime with this condition, was heard by the police officer. The Police officer did not have any discretion. The system set up was such that once the AI face recognition identifies the suspect, Police are required to arrest her, bring her to the police station and book her.

In this case, she was taken to the police station, booked and released on bail. Few days later the case against her was dismissed as the AI system has wrongly identified her. It was also found out that she was not the first case and few more people, especially African American women were wrongly arrested and released later due to incorrect facial recognition model.

The speed in which the governments are moving on regulations and proliferation of AI tech companies delivering business application such as this facial recognition model demand urgent regulations.

May be citizens themselves should organize and let the people responsible for deploying these systems accountable. The Chief of Police, may be the Mayor of the town and County officials who signed off this AI facial recognition system, should be made accountable. May be the County should pay hefty fines and just not a simple oops, sorry.

Lots of attention need to be placed on training data. Training data should represent all the diverse people in the country in sufficient samples. Expected biases due to lack of sufficient diversity in training data must be anticipated and the model tweaked. Most democratic countries have criminal justice system with a unwritten motto “Let 1000 criminals go free but not a single innocent person should go to jail”. The burden of proof of guilt is always on the state. However, we seem to have forgotten this when deploying these law enforcement systems. The burden of proof with very high confidence levels and explainable AI human understandable reasoning, must be the basic approval criteria for these systems to be deployed.

The proposed EU act classifies these law enforcement systems as high risk and will be under the act. Hopefully the EU act becomes a law soon and avoid this unfortunate violation of civil liberty and human rights.

More Later,

L Ravichandran

EU AI Regulations Update

Posted on August 18, 2023 - 5:30 am by L Ravichandran

I have written some time back about EU AI Act draft circulation. After more than 2 years, there is some more movement in making this a EU Law. In June 2023, the EU Parliament adapted the draft and a set of negotiating principles and the next step of discussions with member countries has started. The EU officials are confident that this process will be completed by end of 2023 and this will become an EU law soon. Like the old Hindi proverb “ Bhagawan Ghar mein Dher hain Andher Nahin”. Or “In God’s scheme of things, there may be delays but never darkness”. EU has taken the first step and if this becomes a law by early 2024, it will be a big achievement. I am sure USA and other large countries will follow soon.

The draft has more or less maintained its basic principles and structure.

The basic objective of the new law is to make sure that AI systems used in the EU are safe, transparent, traceable, non-discriminatory and environmentally friendly. In addition, there is an larger emphasis on AI systems should be overseen by people, rather than by automation alone. The principle of proportionate regulations, the risk categorization of AI systems and the level of regulations appropriate to the risk are the central theme of the proposed laws. In addition, there was no generative AI or ChatGPT like products when the original draft was developed in 2021 and hence additional regulations are added to address this large language models / Generative AI models. The draft also plans to establish a technology-neutral, uniform definition for AI that could be applied to future AI systems.

Just to recall from my earlier Blog, the risks are categorized in to Limited risk, high risk and unacceptable risk.

The draft Law clearly defines systems which are categorized as “Unacceptable risk” and proposed to ban them from commercial launch within EU community countries. Some examples are given below.

Any AI system which can change or manipulate Cognitive behaviour of humans , especially vulnerable groups such as children, elderly etc.

Any AI system which classifies people based on various personal traits such as behaviour, socio-economic stataus or race and other personal characteristics.
Any AI system which does real-time and remote biometric identification systems, such as facial recognition which is usually without consent of the person targeted. The law also clarifies that past data analysis for law enforcement purposes is acceptable with court orders.

The draft law is concerned about any negative impact on fundamental rights of EU citizens and any impact on personal safety. These types of systems will be categorized as High Risk.

1) Many products such as toys, automobiles, aviation products, medical devices etc. are already under existing U Product safety legislation. Any AI systems that are used inside products already regulated under this legislation will also be subjected to additional regulations as per High Risk category.

2) Other AI systems falling into eight specific areas that will be classified as High Risk and required registration in an EU database and subjected to the new regulations.

The eight areas are: –

Biometric identification and categorisation of natural persons
Management and operation of critical infrastructure
Education and vocational training
Employment, worker management and access to self-employment
Access to and enjoyment of essential private services and public services and benefits
Law enforcement
Migration, asylum and border control management
Assistance in legal interpretation and application of the law.

Once these systems are registered in the EU database, they will be assessed by appropriate agencies for functionality, safety features, transparency, grievance mechanisms for appeal etc and will be given approvals before they are deployed in EU market. All updates and new versions of these AI system will be subjected to similar scrutiny.

Other AI systems not in the above two lists will be termed as “Limited risk” systems and subjected to self-regulations. At the minimum, the law expects these systems to inform the users that they are indeed interacting with an AI system and provide options to change to a human operated system or discontinue using the system.

As I have mentioned before, the proposed law is covering Generative AI systems also. The law required these systems to disclose to the users that the output document or a output decision is generated or derived by a Generative AI system. In addition, the system should publish the list of copyrighted training content used by the model. I am not sure how practical this is given that ChatGPT like systems are reading every digital content in the web and now moving in to very audio / video content. Even if the system produces this list which is expected to be very large, not sure current copy right laws are sufficient to address the use of this copyrighted material in a different form inside the deep learning neural networks.

The proposed law also wants to ensure that the generative AI models are self-regulated enough not to generate illegal content or provide illegal advice to users.

Indian Government is also looking at enacting AI regulations soon. June 9^th 2023 interview, Indian IT minister talked about this. He emphasized the objective of “No harm” to citizen digital users. Government’s approach to any regulation of AI will be thru the prism of “ User harm or derived user harm thru use of any AI technology”. I am sure draft will be out soon and India also will have similar laws soon.

Let us discuss about what are the implications or consequences of this regulation among the various stakeholders.

AI system developer company ( Tech and Enterprises )

They need to educate all their AI development teams on these laws and ensure these systems are tested for compliance prior to commercial release. Large enterprises may even ask large scale model developers like open.AI to indemnify them against any violations while using their APIs. Internal legal counsels of both the tech companies and API user enterprises need to be trained on the new laws and get ready for contract negotiations. Systems Integrators and outsourcers such as Tech Mahindra, TCS, Infosys etc. are also need to gear up for the challenge. The liability will be passed down from the enterprise to the Systems Integrators and they need to ensure compliance is built in and also tested correctly with proper documentation.

Governments & Regulators

Government and regulatory bodies need to upskill their staff on the new laws and how to verify and test compliance for the commercial launch approval. The tech companies are very big and throw in best technical as well as legal talent to justify their systems are compliant and if regulatory bodies are not skilled enough to verify then the law will become ineffective and will be only on paper. This is a huge challenge for the government bodies.

Legal community both public prosecutors, company legal counsels and defence lawyers

Are they ready for the avalanche of legal cases starting from regulatory approvals and appeals, ongoing copyright violations, privacy violations, inter company litigations of liability sharing between Tech, enterprise and Systems Integrators etc.

Massive upskillng and training is needed for even senior lawyers as issues arising from this law are very different. The law degree curriculum needs to include a course on AI regulations. For example, the essence of a comedian talk show “learnt” by a deep learning model and stored deep in to neural networks. Is it a copyright violation? The model outputs similar style comedy speech by using the “essence” stored in neural network. Is the output a copy right violation? Who is responsible and accountable for an autonomous car accident? Who is responsible for a factory accident, causing injury to a worker in a autonomous robot factory? Lots of new legal challenges.

Most Indian Systems Integrators are investing large sums of money to reskill and also create new AI based service offerings. Hope they are spending part of that investment in AI regulations and compliance. Otherwise, they run a risk of losing all the profits in few tricky legal challenges.

More later

L Ravichandran

What NOT to say

Posted on February 7, 2022 - 5:17 am by L Ravichandran

What Not to Say

Teaching chatbots to speak ‘properly’ and ‘decently’

Many of us would have heard about Microsoft’s Tay.ai chatbot, which was released and pulled back within 24 hours in 2016, due to abusive learnings by the chatbot. It took less than 24 hours to corrupt an innocent AI chatbot. What went wrong? Tay.ai’s learning module was excellent, which ironically was the problem – it was rapidly learning swear words, hate language etc. from the large number of people who used abusive language during conversations with the chatbot. However, unlike some of the internal filters many of us have, Tay.ai went ahead and learnt from these signals, and started using these phrases and hate language. All this happened in less than 24 hours, which forced Microsoft to pull this from public use.

I have been observing how my son and daughter-in-law are teaching my 3-year-old granddaughter about the use of good language. Basic things like saying ‘Please’, ‘Thank You’, ‘Good morning’, ‘Good night’, etc. In other words, decent and desirable language was taught first. They have also given strict instructions to us (grandparents) and extended family about what to say – and what not to say – in front of the kid. The child will still hear some ‘bad words’ in school, malls, playgrounds etc. This is beyond the parents’ control. In these cases, they teach the child about how a very few bad people still use ‘bad’ language and good people never use these words, thus starting to lay in the internal filters in my granddaughter’s mind.

We should apply the same principle to these innocent but fast-learning chatbots. Let us ‘teach’ the chatbot all the ‘good’ phrases like ‘Please’, ‘Thank you’ etc. Let us also ‘teach’ the chatbot about showing empathy, such as saying ‘Sorry that your product is not working. We will do everything possible to fix it’, ‘Sorry to ask you to repeat as I did not understand your question’, and so on.

Finally, let us create a negative list of ‘bad’ phrases, and hate language in all possible variations. English in the UK will have British, Scottish, and Irish variations. Some phrases which are considered acceptable in one area may be objectionable in another. Same for Australia, New Zealand, India, New York Northern English, Southern USA English, etc. Let us build internal filters in these chatbots to ignore or unlearn these phrases in the learning process. By looking at the IP address of the user, the bot can identify the geographical location and apply the right language filters.

Will this work? As good parents we have been doing this to teach our kids and grandkids from time immemorial. Mostly this is working; very few kids grow to become users of hate language.

Will it slow down the machine learning process? Perhaps a little bit, but this is a price worth paying, compared to having a chatbot use foul language and upset your valuable customers.

You may be wondering if this simple approach is supported by any AI research or whether this is just a grandfather’s tale! There is lots of research in this area that supports my approach.

There are many references to articles on ‘Seldonain Algorithm’ for AI Ethics. I want to refer to an article titled ‘Developing safer machine learning algorithms at UMass Amhrest’. The authors recommend that the burden of ensuring that ML systems are well-behaved is with the ML designer and not with the end user, and they suggested a 3-step Seldonian algorithm. Let us look at this.

Step one is to provide an Interface specified by the user to define undesirable or bad behaviour. The ML algorithm will use the interface and try as much as possible to avoid these undesirable behaviours.

Step two is to use High-Probability Constraints: Seldonian algorithms guarantee with high-probability that they will not cause the undesirable behaviour that the user specified via the interface.

Step three in the algorithm is No Solution Found: Seldonian algorithms must have the ability to say No Solution Found (NSF) to indicate that they were unable to achieve what they were asked.

Let us consider two examples involving human life to illustrate the Interface definitions. Example one is a robot that controls a robotic assembly line. The robot senses that a welding operation has gone out of sync and is causing all welded cars to be defective. The robot controller wants to issue the instruction to immediately stop the assemble line and get the welding station fixed. However, the user knows that abrupt stoppage of assembly line may cause harm to some factory workers who may be on another station in the assembly line. This undesirable decision to immediately stop the assembly line needs to be defined in the interface, as this will cause harm to humans compared to a material loss in defective cars.

Example two is an autonomous truck carrying cargo driving in a hilly road with a cliff on the driving side. A human driver is coming fast in the wrong lane ( human’s fault) and approaching the truck for a certain head-on collision. The only desirable outcome for the truck is to fall of the cliff and destroy itself with the cargo rather than trying to look at various other optimal decisions which may have some probability of hitting the car and harming the human.

In our chatbot good-behavior problem, the undesirable behaviors are usage of the phrases in the ‘Negative List’ for each geographical variation. The interface will have this list and the logic to identify geographical variations.

I am in discussion with some sponsors for a research project to develop an English-language chatbot etiquette engine. Initial reactions from the various stakeholders are positive – everyone agrees on the need for an etiquette engine as well as my approach.

I will be delighted to receive critique and comments from all of you.

As a closing note, wanted to tell you that Natural Language processing (NLP) is taking huge strides. “NLP is eating the ML” is the talk of the town. NLP research supported by Large Language models, Transformers etc. are moving way ahead. Investment is going into Q&A, Language Generation, Knowledge management, Unsupervised/reinforcement learning.

In addition to desirable behavior, many other ethical issues need to be incorporated. For e.g

· Transparency: Does everyone know broadly how learning is done and how decisions are taken?

· Explainability: For every individual decision, if requested, can we explain how the decision was taken?

Also, a lot of current AI/ML algorithms especially neural networks based have become black boxes. We expect a shift towards more simpler algorithms for enterprise usage.