Generative AI - Turning Options Into Action With “The Five Ps”.

2023 saw an explosion of AI models, wrappers, and capabilities. A cornucopia of cyber possibilities and offerings suddenly bloomed. There will be no let-up. Ever since Google developed the Transformer model in 2017, a neural network that learns context and generates new content by understanding relationships in sequential data, like the words in this sentence, Generative Artificial Intelligence (Gen AI) has seen successive step changes in its computational and memory power derived from unfathomably large training data sets. Huge changes in the quantity of data have resulted in major qualitative shifts in generative output, and they continue to do so at accelerating speed since OpenAI’s release of ChatGPT into the public domain in November 2022.

Traditional AI is based upon predefined algorithms and produces deterministic outcomes from logical problem-solving and decision-making. For example, think of pre-programmed searching of a database of information for the most appropriate answers to customer service questions. Gen AI, however, is fundamentally different. It is far more impactful and versatile, representing a complete shift in capability by harnessing what are known as Deep Learning techniques (a subset of Machine Learning in the realm of Artificial Intelligence). Because Gen AI learns and generates new content from the underlying patterns and structures of the information and language it ingests, its output is not entirely predictable. Indeed, Generative AI is constantly evolving and in ways that, while incredibly useful to us, we simply do not fully understand.

The market is dominated by the key Foundation models of Generative AI, as they are known, including OpenAI’s ChatGPT, Google’s Gemini, Meta’s open-source Llama, and others. Whether you’re looking for computer code, policy ideas, translation, drafting emails, an employee manual interlocutor, transcription, marketing insights, synthesized analysis, or a handy synopsis, Gen AI literally does it all.

The industry continues to work on the development of safeguards and guardrails against Gen AI’s many well-documented risks, such as inaccuracies, bias, intellectual property violations, and hallucinations (making things up). Writ large, Gen AI is moving forward relentlessly and exponentially in what has become a geopolitical cyber-arms race. And all this in the space of what is, in relative terms, a nanosecond. It was only yesterday that the technology industry and corporate users were preoccupied with accumulating reservoirs of data seemingly at their own pace. This was the culmination of a process that began in the early 1970s that has been termed The Age of Information.

Now, suddenly, corporate attention has turned to the immediate need to leverage data as effectively and as efficiently as possible, at a pace of technological change never experienced, while bringing the rest of the world along for the ride. The Age of Synthesis is upon us: the synthesis of Gen AI with an incalculable number of human activities and interactions due to its versatility, the synthesis of risk and reward in a way not previously seen due to the agentic output of Gen AI, and the synthesis of multi-disciplinary teams to manage that output. And as this momentous shift occurs, the luxury of sitting on vast amounts of data has already disappeared, with newfound computational and output capabilities that surpass the limits of the human imagination.

So, what does all this mean for your business? Competition may be compelling you to act, but you would be forgiven for asking exactly how this technology is supposed to work for you. Which tools should you use, and for what purposes? You may have some ideas, but the devil is in the details. I echo that. Thought first needs to be given to reviewing the landscape of potential uses within your organization and evaluating their utility. You need to consider the demands of your market, innovation, your readiness, staff resources, the investment required, and, of course, returns.

There is an old maxim that I think often comes in handy (known as the Five Ps): Proper Preparation Prevents Poor Performance. The consultants McKinsey, in their 2023 round-up survey on Gen AI indicated high levels of unpreparedness for the seismic shifts ahead. Nevertheless, “three-quarters of all respondents expect gen AI to cause significant or disruptive change in their industry’s competition in the next three years.”

The results of these changes will be universal, if uneven, but there are many functions and sectors, such as marketing, knowledge-based consulting, and human resources functions, which will be disproportionately affected by this technological tornado.

So, how should you go about successful deployment? I have an alternative proposition for the Five Ps that, I believe, is specifically applicable to this arena, that it seems to me encapsulates an effective approach: Policy, Pathways, People, Procedures, and Platform. This is where you get to think about your use cases for Gen AI and develop a plan to suit your objectives. A plan that allows you to pursue your goals with flexibility and assurance at a speed and in a direction that makes sense for you.

Policy

The first issue to consider, as for all strategies, is framing. Unlike many tech solutions, Gen AI has the potential to touch upon every corner of your business. As such, from the Board to Cybersecurity to the IT group to every employee engaged in its use, it is vital to implement a robust and sensible policy framework (see an example for download here) coupled with ongoing training and review. CTOs and CIOs need to identify and manage its associated risks, including data security, compliance, and biased or inaccurate output. They need to implement ongoing assessments and mitigation strategies. Model selection, technical integration, rigorous - and traceable - procedures along with continuous user training are all critical. Tech team leaders must ensure that Gen AI models are regularly audited and comply with regulations - which may also include those of other jurisdictions in which you do business or where you have customers.

Pathways

Okay, so now what? You have a policy. What are the action steps and what is important in determining which Gen AI models to use? Firstly, the safety and security of any models deployed must be a priority. All Gen AI models being considered should be evaluated by tech team leaders for accuracy, fairness, ethics, bias, security and, indeed, cost.

Some of the issues around Gen AI technology selection are not new. Vulnerability to data breach is an obvious red flag. Other issues take you into new territory. Given the ubiquity of Gen AI, how will you discourage unauthorized access of non-sanctioned models by your employees? How will you ensure consistency in their manner of use by authorized employees? How will you build an audit trail to monitor the performance of these tools? How will you ensure that any negative issues are dealt with and/or officially reported on? Theory is one thing – mapping out the steps and making it happen is quite another.

Beyond these issues, model selection will vary depending upon your specific use cases. Are they internal, market-facing, or both? There are different modalities (types of media handled) with different areas of focus and multiple models available. LLMs for research, text-to-video in real-time, cross-functional analysis, process documentation, human resources management, innovative design, and multi-media marketing are all at your fingertips. Which modalities make sense in your business? What are your goals? What is the optimal combination of models to deliver on these needs? Will you need to chain them together in a sequence of operations? These are all questions to think about.

You may opt for models to be hosted within your firewall for reasons of greater security and internal use case needs. Alternatively, you may prefer to use cloud-based models for marketing-based initiatives or other approaches. Either way, you will need to consider workflow integration, independence from model vendor lock-in, and ease of integration into your existing tech stack.

People

Which brings us to your all-important asset - people. After all, these systems will not operate themselves. Or, at least, they shouldn’t. The automation function of Gen AI is well documented, but review is critical. It also makes sense to conduct a detailed impact assessment and review of specific staff functions to understand more fully the enhancement capabilities of Gen AI, and its potential for reorganizing teams while driving the top and bottom lines.

In time, Gen AI will turn many traditional staffing models on their head. Teams will need to be responsive to change in an extremely dynamic environment and be willing to adopt new business practices. According to Forbes Magazine in a recent study, between 70% and 95% of enterprises are currently failing to make the grade in their business transformation process. The skill gaps with existing personnel are cited as the primary reason for these failures. Given the novelty of the technology, much of this will come down to experimentation and hands-on learning in a training environment.

Thought may need to be given to whether a Chief of Artificial Intelligence post is warranted, with a hub-and-spoke system of Gen AI coordinators supporting your various departments. This may also facilitate multi-disciplinary collaboration on various projects across departments. One thing is for sure: the output of Gen AI itself will, at least in part, dictate the pace on this front. Synthesized output from Gen AI compels new team configurations. If silos once seemed outdated and inefficient, they have now become untenable.

Teams will undergo radical transformation. Leaders should support their teams in rapid upskilling to keep pace with the changes underway. An internal forum for sharing best practices may help to remove obstacles and create leverage. Initially, it makes sense to conduct a skills-gap analysis while setting strategic objectives and determining the right models to use. At the outset and as Gen AI becomes more integrated into workflows, leaders should facilitate the training and upskilling of their teams, including educating team members on AI ethics and responsible AI practices.

Procedures

First up – your data. There are many types. For example, customer data, customer interaction data, transaction data, financial performance data, operational performance data, etc. Is the data in good shape? Structured or - let’s be charitable - unstructured. Appropriately sanitized and usable without risk of privacy violations? How do you intend the data to be used, and by which models? Specific usage procedures around prompt requests (the questions asked of Gen AI) will also need to be developed. Perhaps shared libraries will be useful. Structured prompts may be developed over time. Certainly, you will need to create space for experimental development. More things to think about. I will explore prompting in more detail in a future article.

Then there are the expectations of responsible use by employees. Realistically, while some organizations attempt to do so, you cannot completely prevent the use of Gen AI tools with a blanket moratorium. In fact, the statistics for unauthorized use of Gen AI on company business are staggering, with one survey indicating two-thirds of use is unauthorized. What you can and should do, however, is establish guidelines for the use of Gen AI within your organization (back to Policy and communication). Employees should also document the prompts they use to generate results, proofread Gen AI output, and sign-off protocols should be in effect, with peer-review where required. This is what is known as the human-in-the-loop requirement. It is imperative. Data anonymization is also important to ensure that sensitive information is not unwittingly shared with models, especially where retrievability is practically impossible. All of this will require training, as previously noted.

Platform

When you put all these pieces together, the picture that emerges of Gen AI is one of a fundamental and complex operation of powerful agents at work within your organization. This web of interactions needs to be managed effectively, as it is likely to become all-pervasive in the way that the world does business. Within your organization, how will you extract the benefits of Gen AI whilst also establishing monitoring procedures that incentivize your teams? It seems an insurmountable task in the absence of some form of central tracking platform, if you wish both to gauge increased productivity and to mitigate and report on any emerging risks. Compliance will ultimately demand this.

Establishing a governing Platform - a permission-based software architecture - that makes Policy and Procedure concrete based upon the Pathways that your People are clear on, is the only viable way forward.

“The Five Ps” - turning options into action.