Since ChatGPT was released in November 2022, it has created a lot of buzz in tech circles, but also in the mainstream. It’s prompted companies to race to market their latest (and usually greatest, though that’s starting to change) LLMs, from Microsoft to AWS, Google to Facebook and a host of open source models. Those experimenting with ChatGPT range from middle school teachers to company CEOs, and the combined momentum has generated a lot of buzz and concern about the potential for benefits and harm.
Criticism intensified last week due to a series of events. Last Wednesday, AI experts, industry leaders, researchers and others signed an open letter calling for a six-month “pause” on large-scale AI development outside of OpenAI’s GPT-4. The next day, the AI ​​policy group CAIDP (Center for AI and Digital Policy) filed a complaint with the FTC, arguing that the use of AI should be “transparent, explainable, fair, and empirically sound while promoting accountability” and that OpenAI GPT of -4 “meets none of these requirements” and is “biased, deceptive, and a threat to privacy and public safety.” Finally, on Friday, the first major chip to drop was Italian regulators’ call on OpenAI to address certain privacy and access concerns surrounding ChatGPT.
As an AI company providing natural language (NL) solutions founded in Italy, this development was of particular interest. At expert.ai, our approach does not depend on LLM (large language model). In fact, we propose combined approaches through hybrid NL, which allows for transparency. This is one of our core principles and one of the four aspects of our “green glass” approach to responsible AI.
Why did Italy “ban” ChatGPT?
order issued by The Italian Data Protection Authority (Garante per la protezione dei dati personali) notes personal data protection concerns related to the GDPR, namely that ChatGPT processed personal data unlawfully and that there is no system in place to prevent access by minors. OpenAI has disabled ChatGPT in Italy. It has 20 days to respond to the order, and if it doesn’t, it could face significant financial penalties of up to 4% of annual turnover or €20 million. The specific issues mentioned are on the list of key considerations businesses should be aware of. which we highlighted back in February.
- ChatGPT was trained on data scraped from the Internet, including articles, social media content, Wikipedia, and Reddit forums. As the TechCrunch article points out, “if you’ve been online for a reasonable amount of time, chances are the bot knows your name.”
- ChatGPT is generative AI, meaning it creates human language content based on predicting words and sentences that might come next based on its training. This makes it vulnerable to “hasty injections” that can be used to deliberately manipulate the content it produces, with potentially dangerous effects.
- ChatGPT is not based on understanding the relationships and context within the language. While it can generate content in human language that looks and sounds right, it can also articulate things quite convincingly, which is called “hallucination”.
- The LLM that ChatGPT is based on is considered a “black box” transformer that is not interpretable, meaning you cannot track how it produced the results (whether accurate or not).
How does this affect the future of ChatGPT and LLMs in general? What do businesses need to know about deploying their own LLM?
Considerations for AI applications moving forward
The concerns raised are a reminder of the risks of some types of AI, but it’s also time to address the proven capabilities of AI already in place. At expert.ai, we’ve delivered over 300 natural language AI solutions over the past 25 years, working with many Fortune 2000 companies to optimize processes and improve people’s performance. We don’t just insist on being human, we work to humanize the work done to make it more engaging and add value to people’s solutions.
In that regard, we want to share some general considerations for using any AI to solve your real business problems.
1. Transparency and explainability must be built into any AI solution
Large language models such as GPT-3 and GPT-4 are so large and complex that it is the ultimate “black box AI” approach. If you can’t determine how the AI ​​model arrived at a certain decision, it can end up being a business problem and, as we’re seeing now, a regulatory problem. It is absolutely critical that the AI ​​you choose can deliver results that are easily explainable and accountable.
The way for artificial intelligence to solve real-world problems with the highest accuracy is through a hybrid approach that combines different techniques to take advantage of the best of all worlds.
Symbolic techniques apply rules, and in the case of expert.ai, a rich knowledge graph—all elements that are fully audible and understandable to humans. When combined with machine learning or LLMs, these combined techniques introduce much-needed transparency into the model, providing a clear view of how the system behaves in a particular way to reveal potential performance issues, security concerns, biases, and more.
2. The data you use matters
The data you choose to train any AI model is important, whether you’re working with LLM, machine learning algorithms, or some other model.
Public domain data, such as the data used to train ChatGPT, is not enterprise-level data. Even though ChatGPT’s content spans multiple domains, it does not represent what is used in the most complex enterprise use cases, whether vertical domains (Financial Services, Insurance, LifeSciences, and Healthcare) or highly specific use cases (contract review, medical claims, risk management). assessment and cyber policy review). So even for chat/search use cases that work like ChatGPT, it will be quite difficult to have quality and consistent performance on highly specific domains.
As we mentioned in our previous post, the nature of the data that ChatGPT was trained on also raises concerns for copyright infringement, data privacy, and the use and disclosure of personally identifiable information (PII). Here it contrasts with the European Union’s GDPR and other consumer protection laws.
Natural language AI is most useful when it builds, augments, and captures domain knowledge in an iterative fashion. This requires engineering guardrails (like the combination of AI approaches we use at Hybrid NL), embedded knowledge, and people in the loop. We built our platform on all three of these pillars for that reason, and because it enables businesses to create proactive and sustainable competitive advantages with their AI tools.
3. A people-centered approach is extremely important
Having humans only at the beginning or end of an AI process is not enough to ensure accuracy, transparency, or accountability. Enterprises need a human-centric approach where data and inputs can be controlled and refined by users throughout the process.
Explainable-only and explainable-by-design AI models offer full human control during the development and training phases. Because it involves an open, interpretable set of symbolic rules, hybrid AI can offer a simpler way to correct underperformance. So if the result is misleading, biased, or wrong, users can intervene to prevent future errors and achieve the success criteria most valuable for each use case and improve accuracy, all while keeping the human subject matter expert informed.
While black-box machine learning models only allow more data to be added to the training set without the ability to interpret the results, a hybrid approach can incorporate linguistic rules to handle model data. Hybrid AI enables humans to control the model.
The end result should also include an analysis of not only the ROI provided, but also the human benefit. Were trivial or redundant tasks automated, and does the solution have value for the people receiving the benefits? While ROI, consistency, and automation are typical outcomes in any AI project, natural language solutions, including natural language solutions, often provide additional upside. The work that people do is more engaging, critical and rewarding. Combined with humans, this is human-centric AI.
Looking ahead
As the use of AI increasingly grows into the mainstream, organizations are looking for solutions that can keep them competitive while delivering business benefits, regulatory compliance and internal accountability.
Like any new technology, businesses need to be able to understand how to apply it to solve real problems that don’t put their businesses at risk.
The GPT and other LLMs present real opportunities, but also require a careful focus on administration, accuracy, bias, and cost. As companies experiment with ways to commercialize these technologies, we believe that incorporating knowledge-based approaches provides an important tool for delivering accurate business results in a practical and responsible manner.
Our enterprise AI platform is built around this approach: to provide tools, workflows, and hybrid AI approaches to solve real-world problems in the most responsive, cost-effective way possible.
We offer a management framework to find the best solutions that bring value from language data. We have always promoted our “green glass” definition of Responsible AI, which provides solutions that are transparent, sustainable, practical and human-centered.