Introduction: Generative AI is widely considered the fastest moving technology innovation in history. It has captured the imagination of consumers and enterprises across the globe, spawning incredible innovation and along with it a mutating market ecosystem. Generative AI has also caused a copious amount of FOMO, missteps, and false starts. These are the classic signals of technology disruption—lots of innovation but also lots of mistakes. It is a rumpus room with a lot of “kids” going wild. The rumpus room needs adults. Guidance through the generative AI minefield will come from thoughtful organizations who do not panic, who understand the fundamentals of AI, and who manage risk.
2023 was essentially the first year of the generative AI era. To say 2023 was a crazy year in AI would be a gross understatement, but in retrospect, the generative AI rumpus room got gradually calmer as the year progressed, particularly in the last 4 months of the year. Since August 21, The Futurum Group has published 10 Adults in the Generative AI Rumpus Room notes, highlighting 30 different adult initiatives from a total of 21 different companies. I can draw some conclusions from that body of work to deliver what we see as the best of the Adults in the Generative AI Rumpus Room for 2023, including:
- Top “adult” initiatives: Initiatives I noted that had the greatest impact in providing adult leadership in generative AI in 2023.
- Top “adult” trends: In reviewing the collective initiatives, what trends emerged in 2023?
- Top adults: Which companies were the most adult in 2023, based on the number of initiatives I noted?
Top Adult Initiatives of 2023
All 30 initiatives I have written about in 2023 are having a positive impact in calming the generative AI market, but these seven will likely make the biggest difference (listed in no particular order):
- Salesforce commits to the trusted use of AI
- IBM commits to list source data for new Granite AI models
- AI2 debuts open dataset for AI training
- Data Provenance Organization launches audit tool for AI datasets
- Google launches a sensitive data protection service and generative AI
- IBM watsonx.governance tackles AI risk management
- Guardrails for Amazon Bedrock levels up responsible AI
Salesforce Commits to the Trusted Use of AI
The News: At Dreamforce 2023 on September 12, CEO Marc Benioff introduced the company’s Tenets of Trusted, Ethical and Humane AI and emphasized Salesforces’ commitment to the trusted use of AI.
Salesforce’s Tenets of Trusted, Ethical and Humane AI:
- Your data is not our product.
- You control access to your data.
- We prioritize accurate, verifiable results.
- Our product policies protect human rights.
- We advance responsible AI globally.
- Transparency builds trust.
Throughout his keynote, Benioff spoke over and over again about AI, and in the same sentence, trust. He said you cannot do AI without being able to trust what it does. The company is liberally using the word trust in describing AI capabilities. He said we have an incredible opportunity in AI. How we do it, matters. We have to do it right. We have to do it responsibly.
Watch the Dreamforce 2023 main keynote on the Salesforce website.
Adults because… The Dreamforce statements were the latest in a string of responsible actions Salesforce has taken with AI. There is a lot of public conversation about regulation when it comes to privacy and bias and toxicity and hallucinations, transparency, explanability, consumer rights, intellectual property (IP) and copyright laws, consumer protection, and how those regulations could squash innovation and business growth and US competitive advantages.
Maybe that discussion should have a different arc. A conversation that is less about what specifically we should do about government regulation and more about what companies can do to act responsibly with AI right now, and in that process, calm the nerves of all parties—AI users, companies leveraging AI, and government.
What is missing in the equation is representation from a company that is on the front line of live AI application, one that is taking the risk of leveraging AI (and reaping some reward). Further, Washington needs to hear about what that experience has taught such a company, and that implementing the pillars of responsible AI (Responsible, Accountable, Transparent, Empowering, Inclusive, Salesforce Trusted AI Principles) is good business discipline, which mitigates AI risk for both the company offering the service and its customers. Salesforce is living what enterprises need to see in how to incorporate AI responsibly.
IBM Commits to List Source Data for New Granite AI Models
The News: On September 7, IBM announced several AI-focused rollouts and enhancements to watsonx, including the introduction of IBM’s own AI models, Granite. The Granite series are IBM-built AI foundation models that the press release says are designed to support enterprise natural language processing (NLP) tasks such as summarization, content generation, and insight extraction. Granite models were made available later in September.
Read the full press release on the Granite AI model and watsonx enhancements on the IBM website.
Adults because… IBM plans to “provide a list of the sources of data as well as a description of the data processing and filtering steps that were performed to produce the training data for the Granite series of models.” There is growing momentum for this approach to transparency in AI models. Many proprietary large language models (LLMs) refuse to divulge their training data sources, for various reasons, though most commonly because the LLM vendor sees that data source as competitive IP. IBM is positioning its AI models less as secret sauce; rather, the proprietary value in the IBM stack is in watsonx and the value of the complete chain. It frees up IBM to offer AI models that might meet transparency best practices.
AI2 Debuts Open Dataset for AI Training
The News: On August 18, the announced the availability of Dolma, a dataset of 3 trillion tokens. It is the largest open dataset to date. Dolma is the dataset AI2’s planned open source LLM, OLMo, will be based on. Nearly all datasets on which current LLMs are trained on are private.
Read the details of Dolma on the AI2 blog.
Adults because… Most LLMs have been built on datasets that are private. The data is typically scraped, without permission, from publicly available data on the web. The major challenges of LLM outputs include bias, toxicity, inaccuracy, and hallucination. One way to address these issues is for those who use the LLMs to be able to trace these issues back to the data source. Open datasets provide that opportunity.
Data Provenance Organization Launches Audit Tool for AI Datasets
The News: On October 25, a newly formed, researcher-led group, the Data Provenance Organization, published a paper and data that will enable organizations to audit AI datasets used to train LLMs. The abstract is compelling:
“The race to train language models on vast, diverse, and inconsistently documented datasets has raised pressing concerns about the legal and ethical risks… we convene a multi-disciplinary effort between legal and machine learning experts to systematically audit and trace 1800+ finetuning datasets. Our landscape analysis highlights the sharp divides in composition and focus of commercially open vs closed datasets, with closed datasets monopolizing important categories: lower resource languages, more creative tasks, richer topic variety, newer and more synthetic training data. This points to a deepening divide in the types of data that are made available under different license conditions, and heightened implications for jurisdictional legal interpretations of copyright and fair use. We also observe frequent mis categorization of licenses on widely used dataset hosting sites, with license omission of 72%+ and error rates of 50%+. This points to a crisis in misattribution, and informed use of the most popular datasets, driving many recent breakthroughs. As a contribution to ongoing improvements in dataset transparency and responsible use, we release our entire audit, with an interactive UI, the Data Provenance Explorer, which allows practitioners to trace and filter on data provenance for the most popular open source finetuning data collections.”
Adults because…The old adage is garbage in, garbage out, and it applies to large datasets used to train AI. Using better (and usually smaller) datasets will lead to better LLM outcomes—more accurate, less misinformation and disinformation. The Data Provenance Explorer gives enterprises the opportunity to review different model’s approaches and “look under the hood” at the data, or at least, how it is constructed, to be better informed about the models they might choose to use. Increased use of such audits could force model developers to fine-tune their datasets.
Google Launches a Sensitive Data Protection Service and Generative AI
The News: On October 4, Google Cloud published a post describing how their Sensitive Data Protection service can be used to help secure generative AI workloads. According to the post:
“… generative AI requires data in order to tune or extend it for specific business needs…However, one concern that organizations have is how to reduce the risk of customizing and training models with their own data that may include sensitive elements such as personal information (PI) or personally identifiable information (PII). Often, this personal data is surrounded by context that the model needs so it can function properly.
… Organizations can use Google Cloud’s Sensitive Data Protection to add additional layers of data protection throughout the lifecycle of a generative AI model, from training to tuning to inference. Early adoption of these protection techniques can help ensure that your model workloads are safer, more compliant, and can reduce risk of wasted cost on having to retrain or re-tune later.”
Adults because… To truly unlock proprietary power with an LLM, enterprises must leverage their own data. Most have hesitated to date because of the fear that proprietary data will be used to train models or personally identifiable information (PII) will be exposed. In other words, enterprises have a hard time trusting they can use LLMs and their own data due to security issues. Data security is a major issue for generative AI. Google’s Sensitive Data Protection service is a way for enterprises to feel more confident that they can leverage their own data against AI foundation models.
IBM watsonx.governance Tackles AI Risk Management
The News: On July 11, IBM announced that watsonx.governance, a product designed to enable enterprises to direct, monitor, and manage the AI activities of the organization, is slated to be generally available before the end of 2023. Organizations typically build too many AI models without clarity, monitoring, or cataloging. Watsonx.governance automates and consolidates tools, applications, and platforms. If something changes in models, all the information is automatically collected for audit trail through testing and diagnostics.
Enterprises must understand risks of AI in every use, case by case. Emerging best practices are for organizations to build an AI ethics governance framework, including establishing an ethics committee or board to oversee. This approach is a highly manual process. watsonx.governance addresses this through automating workflows to better detect fairness, bias, and drift. Automated testing ensures compliance to an enterprise’s standards and policies throughout the AI’s lifecycle.
More than 700 AI regulations have been proposed globally. Most are umbrella regulations that are not overly specific. Most companies do not understand how to comply. watsonx.governance addresses this lack of understanding by translating growing regulations into enforceable policies within the company. The solution breaks down the requirements in the regulation and builds controls.
Read the full watsonx GA announcement on the IBM website.
Adults because… AI risk management is foundational and critical to operationalizing AI. Enterprises will learn this either the hard way, through ignoring it, or the easier way, by embracing it. IBM is in a great position to help enterprises navigate AI risk management. The strengths are in the vision and blueprint for automating model management/audit trail and for automating workflows to better detect fairness, bias, and drift and for automating testing throughout the AI lifecycle.
Guardrails for Amazon Bedrock Levels Up Responsible AI
The News: At re:Invent, Amazon Web Services (AWS) launched Guardrails for Amazon Bedrock into preview. With the new tool, Amazon Bedrock users can define denied topics and content filters to remove undesirable and harmful content from interactions between their applications and users. Here are the key details:
- Control: Denied topics and configure with natural language commands. Users can use a short natural language description to define a set of topics that are undesirable in the context of their application.
- Control: Content filters. Users can configure thresholds to filter harmful content across hate, insults, sexual, and violence categories. While many FMs already provide built-in protections to prevent the generation of undesirable and harmful responses, Guardrails gives users additional controls to filter such interactions to desired degrees based on the user’s company’s use cases and responsible AI policies.
Read the AWS new blog post on the launch of Guardrails for Amazon Bedrock on the AWS website.
Adults because… Guardrails for Amazon Bedrock reflects careful thinking by AWS about the responsible use of AI. The prevention/proactive approach is unique at this point, though it is likely that both Microsoft and Google will soon add similar features to their AI development platforms. Regardless, the initiative is the mark of AI leadership and another signal that AWS understands generative AI and is fully engaged in enabling enterprises to leverage generative AI.
For further analysis including comparisons of Guardrails to the Microsoft and Google comparable responsible AI governance tools, read Guardrails for Amazon Bedrock Show AWS Gets Generative AI.
Top Adult Trends of 2023
Building dataset transparency. Three of the top seven initiatives—IBM’s commitment to revealing the sources of the Granite models, AI2 launching the massive open source Dolma dataset, and the Data Provenance Organization’s audit and trace of 1,800 fine-tuning datasets—are focused on building better, more responsible LLMs. The major challenges of LLM outputs include bias, toxicity, inaccuracy, and hallucination. The ways to address these issues is for those who use the LLMs to be able to trace these issues back to the data source.
Emergence of commercial AI risk management tools. IBM’s watsonx.governance and Guardrails for Amazon Bedrock are the first commercial automation tools for AI risk management. The Futurum Group believes that AI governance, risk management, and responsible use of AI are critical for successful implementation of enterprise AI. Look for more emphasis on automation tools and best practice frameworks in 2024.
Top Adults of 2023
Clearly a subjective accolade that could be judged on the importance of the initiatives of particular adults in this analysis or the volume of initiatives an adult was recognized for. For my purposes, the volume of initiatives will be used. Under these criteria, the top adult in the generative AI rumpus room is unequivocable: Google.
#1: Google: Google was responsible for a remarkable six of the 30 adult initiatives tracked, or 20%. The company’s work spanned a range of different areas: Google Android introduces AI safeguard policies for Google Play apps; IP indemnification for generative AI; Sensitive Data Protection service for generative AI; Google Cloud launches SynthID AI watermarking tool; YouTube enlists UMG artists to tinker in YouTube Music AI Incubator; Google advances generative AI search.
#2: IBM: IBM was responsible for three adult initiatives: IP indemnification for generative AI; providing the data sources for the Granite models; and watsonx.governance.
#3 tie: AWS, Anthropic: AWS and Anthropic posted two initiatives each. The complete set of Adults in the Generative AI Rumpus Room notes with details of each initiative are listed below.
Conclusion
All the organizations that stepped up to lead as adults in the generative AI rumpus room in 2023 should be commended. It is likely we are through the most unruly year for generative AI, but 2024 will still require more adults to step up, show the way, and help instill calm and order into the generative AI market.
Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.
Other insights from The Futurum Group:
Adults in the Generative AI Rumpus Room: Anthropic, AWS, Meta
Adults in the Generative AI Rumpus Room: Leica, Data Provenance, Google
Adults in the Generative AI Rumpus Room: Google, Tidalflow, Lakera
Author Information
Mark comes to The Futurum Group from Omdia’s Artificial Intelligence practice, where his focus was on natural language and AI use cases.
Previously, Mark worked as a consultant and analyst providing custom and syndicated qualitative market analysis with an emphasis on mobile technology and identifying trends and opportunities for companies like Syniverse and ABI Research. He has been cited by international media outlets including CNBC, The Wall Street Journal, Bloomberg Businessweek, and CNET. Based in Tampa, Florida, Mark is a veteran market research analyst with 25 years of experience interpreting technology business and holds a Bachelor of Science from the University of Florida.
