What is Responsible AI and how to implement it in your ecommerce product.

Last year was all the rage about generative AI, chatGPT, LLMs, open source models and tools, and investment went thru the roof.

this year, 2024, people came down a bit closer to earth and started asking more questions and challenging assumptions. So the bubble burst a little and it's deflating more as people realize the cost in running LLMs, required for generative AI.

All along there have been people on both sides, some super excited and some not so much warning of the bubble effect and reminding us that AI is not new and has been around for decades.

This is true, AI has been around for decades. What is not correct, is that there is nothing new or special about Generative AI (GenAI) and chatGPT as a revolutionary product.

Specially if you are building consumer-facing app the potential is enormous, to the point of changing the communication models between humans and machines (AI) for good. It is however, wrong to expect things to happen overnight, or to expect perfect products today.

Since at Visional we're in the ecommerce and shopping space for us It's not IF but WHEN the conversational model of communication will take root in ecommerce and shopping.

Nobody has the final answer but most people agree we need to experiment and learn. It would be a mistake to write off GenAI and not even experiment. Even though we're early on in the productization of GenAI for shopping and commerce, the speed of development in tools, models and other aspects is so fast one can easily fall behind to the point of not being able to catch up. At that point catching up will be much more expensive and even impossible.

One of the historical challenges with retail companies has been the lack of experience and understanding technology for typical executive teams. All the way thru the 2010's this was a challenge where CEOs, CFO, even marketing executives knew little about technology and were totally dependent on others to inform them of the impact and the basic questions they need to be asking. The past half decade, specially since the pandemic this dynamic began changing and a lot more retail execs became familiar with technology to hold their own to a reasonable level in conversations.

Executives can fall behind in AI to the point of no return, harming their companies and careers for good.

The same challenge will face retail and e-commerce executive teams in GenAI. If you know nothing, or very little about how AI is changing you're going to not be asking the right questions and always dependent on what others are telling you, and this will be a competitive disadvantage.

Responsible AI

Setting aside the key technical parts of GenAI, and getting to the point of this article, is the concept of Responsible AI. As builders and operators we need to not just think about the technical side of AI, but also of the societal and human side, summed up to the topic of Responsible AI.

This article goes into some basics and how we at Visional are thinking and implementing "responsible" GenAI capabilities into our product.

What is Responsible AI (with help from ChatGPT and other AI tools)

Responsible AI refers to the development and deployment of artificial intelligence systems in ways that prioritize ethical considerations, fairness, transparency, and accountability. It involves creating AI that respects human rights, reduces bias, ensures privacy, and operates in a way that can be understood and trusted by users.

The key principles of Responsible AI typically include:

• Fairness: Ensuring that AI systems do not perpetuate or exacerbate biases and are designed to treat all individuals and groups equitably.
• Transparency: Making AI processes and decision-making understandable and explainable to users and stakeholders.
• Accountability: Establishing clear lines of responsibility for the outcomes of AI systems, ensuring that those who design and deploy AI are held accountable for its impacts.
• Privacy: Protecting individuals' data and ensuring that AI systems comply with data protection laws and best practices.
• Safety and Security: Ensuring that AI systems are robust and secure against threats, and that they operate in ways that do not harm users or society.
• Human-Centricity: Designing AI systems that enhance human well-being, respecting human rights, and ensuring that AI serves humanity rather than undermining it.

Every one of these principals takes careful thinking, experimentation and potential correction to get right. And to top it off, will likely need to be updated at least for the next few years as governments develop new laws and industry builds new guidelines and best practices.

What we aim to do is not try to perfect all of these before launching, but leave the most complex ones for later launches and mitigate them in earlier product releases. This is serious business

Note: Responsible AI is not related to DEI and the conversations around that. As builders we need to take into account how our product can impact our users beyond the direct interactions and usage we intend. There can be side-effect as well as benefits and we should investigate, prepare and implement what can be done as we build.

Applying Responsible AI to commerce

Some of the areas specific to commerce for implementing responsibility into AI include:

1. Fairness in Product Recommendations

• Avoiding Bias in Algorithms: Ensure that your recommendation engine doesn't disproportionately favor certain brands, sellers, or products over others without valid reasons. Regularly test for and mitigate any biases that might favor high-margin items or large retailers over smaller, local businesses.
• Equitable Access: Design the AI to recommend a diverse range of products that cater to different consumer segments, including various price points, brands, and product types.

2. Transparency in Shopping Experiences

• Explainable Recommendations: Clearly communicate why certain products are being recommended. For example, if a shopping agent suggests a product, provide a simple explanation like "Based on your past purchases" or "Trending in your area."
• Clear Pricing Information: Ensure that pricing, discounts, and fees are transparent and easily understood by customers, avoiding hidden costs or misleading promotions.

3. Accountability in Transaction Management

• Handling Complaints and Returns: Create clear processes for handling issues related to incorrect recommendations, errors in product details, or problems with orders. Ensure the AI can quickly escalate issues to human support when needed.
• Audit Trails: Implement logging and monitoring systems that track the AI's decisions, providing an audit trail that can be reviewed if there's a dispute or problem.

4. Privacy in User Data Handling

• Personalized but Secure: Use AI to personalize shopping experiences without compromising user privacy. For instance, anonymize user data when building recommendation models and avoid storing unnecessary personal information.
• Data Usage Transparency: Clearly inform users how their data will be used to personalize their shopping experience, and provide them with options to control what data is collected and used.

5. Safety and Security in Online Transactions

• Fraud Detection: Integrate AI to detect and prevent fraudulent transactions, such as identifying suspicious payment activities or counterfeit product listings.
• Secure Payment Processing: Ensure that the AI shopping agent works within secure, encrypted environments, particularly when handling payment details or personal information.

6. Human-Centricity in User Experience

• User-Centric Recommendations: Tailor product recommendations to the specific needs and preferences of each user, rather than just pushing popular or high-margin products. For example, recommend sustainable products for eco-conscious customers.
• Supporting Informed Decisions: Provide detailed product information, including reviews, ratings, and comparisons, so users can make informed purchasing decisions. The AI should assist rather than manipulate purchasing behavior.

7. Ethical Marketing Practices

• Avoiding Manipulative Tactics: Ensure that the AI doesn’t use manipulative tactics such as creating false scarcity or urgency to push sales. For example, avoid showing “Only 1 left!” messages unless it’s genuinely true.
• Inclusive Promotions: Design AI-driven marketing campaigns that are inclusive and cater to a broad audience, avoiding exclusionary practices or targeting vulnerable groups with unnecessary upselling.

8. Promoting Sustainability

• Eco-Friendly Options: Encourage the AI to prioritize or highlight eco-friendly, sustainable products and vendors. For instance, the shopping agent could flag products with low environmental impact or suggest local options to reduce carbon footprint.
• Waste Reduction: Utilize AI to help consumers avoid over-purchasing, such as by suggesting appropriate quantities or identifying products with a longer shelf life.

The point we most think of are the more direct human-side responsibility effects such as non-biased algorithms, transparency, transactions and processing while other impacts such as sustainability also play a role, specially for the more environmentally conscious customers.

Each of these 8 examples require their own specific data sets as well as technology to inject into our chat feature and although we will use some aspect of our own data and systems it's becoming apparent that we will need external, specialized tools which can go much deeper and wider in each of these areas for our product to work at scale.

We're actively looking at potential products which can help in each of these and if you have a product focusing on one or more of these we'd love to hear from you. Just remember, this is not just putting a skin on an old application, we're looking for products and services built for and on GenAI from ground up.

AI-empowered humans

The concept of empowering humans, sales agents in the store and others in the DC or warehouse is not new. And the benefits are also hard to dispute, assuming cost is reasonable.

GenAI empowered retail staff will perform better in such ways as:

• Better personalization - having quick access to customer history, safely and privately, will result in more engaging interactions and potentially relationships with direct impact on KPI such as LTV and AOV.
• Better product information - GenAI can understand the context behind product description much better than traditional AI, this can lead to reduced search times to find what customers are looking for.
• Better recommendations - understanding customer history, profile and matching it with better product data can result in more relevant recommendation.
• Faster answers to customer questions - we can impress customers or visitors by providing faster responses to their questions.
• More complete answers - more complete answers on product details and other relevant attributes for example weather, seasonal or occasions will save time for both customers and staff and build trust.
• Asking better questions - often it's the questions we ask customers which lead us to the best outcomes. GenAI has the ability to assist staff, or directly, ask better questions/
• Faster transaction processing - all of the above factors can result in faster close times and decision making for consumers.

The application needs to inject AI manually or automatically at the correct points in a conversation and pull in humans when ready. This is an ongoing process for all companies and there is no quick path, excess but getting better at experimentation cycles.

For Visional shopping agents we are building GenAI capabilities into the app so they can better assist and serve customers in all of these aspects. We believe that AG, even the most advances will still not match a caring and experiences human but we can help our agents with technology tools for a better outcome for all.

Data and AI policy governance

The complete cycle of building or adding Generative AI capabilities into e-commerce and retail experiences will require carefully designed and implemented policies and governance. For larger enterprise retailers and apps these need to be addressed in more immediate timelines but newer applications also need to keep these in mind to reduce risk and improve the chances of providing a differentiated experience to customers.

AI and data policies and governance factors in ecommerce are critical to ensure that AI systems are developed and deployed responsibly, with a focus on ethical considerations, regulatory compliance, and user trust.

Here are some key factors to consider:

1. Data Governance

• Data Quality Management: Establish clear guidelines for ensuring the accuracy, completeness, and consistency of data used in AI models. This includes processes for data validation, cleaning, and ongoing monitoring.
• Data Ownership and Stewardship: Define who owns the data collected, how it can be used, and who is responsible for its management. Ensure that data stewardship roles are clearly assigned to oversee data usage, security, and compliance.
• Data Access Control: Implement strict access controls to ensure that only authorized personnel can access sensitive data. Use role-based access to limit data access based on job responsibilities.

2. Privacy and Compliance

• User Consent and Transparency: Ensure that users are fully informed about what data is being collected, how it will be used, and who it will be shared with. Obtain explicit consent from users before collecting their data and provide them with clear options to opt-out.
• Regulatory Compliance: Ensure that your AI and data practices comply with relevant laws and regulations, such as the General Data Protection Regulation (GDPR), California Consumer Privacy Act (CCPA), and other regional data protection laws. Regularly audit your practices to ensure ongoing compliance.
• Data Minimization: Collect only the data that is necessary for the AI system to function effectively. Avoid collecting excessive or irrelevant data, which can increase the risk of privacy breaches and complicate compliance efforts.

3. AI Governance

• Ethical AI Framework: Develop and implement an ethical AI framework that guides the development and deployment of AI systems. This should include principles such as fairness, transparency, accountability, and user-centered design.
• Bias and Fairness Audits: Regularly audit AI models to identify and mitigate biases. This involves testing models against diverse datasets to ensure that recommendations and decisions are fair and do not discriminate against any group.
• Explainability and Transparency: Ensure that AI systems are transparent and explainable, allowing users and stakeholders to understand how decisions are made. This is particularly important in e-commerce, where users need to trust the recommendations and pricing provided by AI.

4. Security and Risk Management

• Data Security: Implement strong security measures to protect data from breaches, including encryption, secure storage, and regular security audits. Ensure that AI systems are designed with security in mind, protecting both data and algorithms from unauthorized access or manipulation.
• Risk Assessment and Management: Conduct regular risk assessments to identify potential threats to the AI system and data. Develop contingency plans to address risks such as data breaches, AI model failures, or compliance violations.
• AI Robustness and Reliability: Ensure that AI systems are robust and reliable, with fail-safes and fallback mechanisms in place to handle unexpected scenarios. Regularly test and update AI models to maintain their accuracy and reliability.

5. Accountability and Oversight

• Clear Accountability Structures: Define clear lines of responsibility for AI and data governance within the organization. This includes assigning roles such as Chief Data Officer (CDO), Chief AI Officer (CAIO), or Data Protection Officer (DPO) to oversee governance efforts.
• Stakeholder Engagement: Engage with stakeholders, including customers, regulators, and industry experts, to gather input on AI and data policies. Ensure that governance practices are aligned with stakeholder expectations and industry best practices.
• Regular Monitoring and Reporting: Implement ongoing monitoring of AI systems and data practices, with regular reporting to leadership and stakeholders. This helps ensure that governance policies are being followed and that any issues are promptly addressed.

6. Training and Awareness

• Employee Training: Provide regular training to employees on AI ethics, data privacy, and security best practices. This ensures that everyone involved in the development and deployment of AI systems understands the importance of responsible AI and data governance.
• User Awareness: Educate users about how AI systems work, what data is being used, and how their privacy is protected. Provide clear communication channels for users to raise concerns or ask questions about AI practices.

7. Continuous Improvement

• Feedback Loops: Establish mechanisms for collecting feedback from users and stakeholders on AI and data practices. Use this feedback to continuously improve governance policies and AI system performance.
• Adaptation to New Regulations: Stay informed about emerging regulations and industry standards related to AI and data governance. Be prepared to adapt your policies and practices to comply with new requirements and best practices.

Companies need tools and expertise to manage Responsible AI and data sides of their business and this will be a growing career track for professionals in ecommerce and retail.

At Visional we are investing in technology and people to build a more responsible experience and product for all our stakeholders - customers, agent, retailer, brands and partners including local chamber of commerce and retail organizations.