Artificial Intelligence (or AI) has meant different things at different times, all through my career. I started working in AI back in the 1990s, when the most prominent use of a neural network was to decode hand-written post code (zip code) digits on letters, and if an organisation was using AI, they had probably implemented an expert system.
This was during an AI winter, when the hype of AI had overtaken expectations, and calling something AI was not considered a positive. Things like the discipline of data science and the technology of speech recognition emerged from this period without being explicitly labelled as AI, and organisations stopped talking about “using AI”.
I worked on implementing intelligent, autonomous agents, and then speech recognition-based personal assistant services. Think of a rudimentary Siri that could arrange meetings by calling people up and asking simple questions about their availability. I also developed a speech-based recommender system that would match people to local restaurants. It didn’t end up going anywhere though.
But AI itself came back in a big way, and organisations started talking about “using AI” when deep learning burst onto the scene in the 2010s. This use of multi-layer neural networks, trained on huge amounts of data with readily-available GPUs, was able to produce results that met or exceeded the results of humans. Seemingly overnight, AI had been redefined to mean deep learning, and all of the data scientists had to wearily explain why their statistical methods should be considered AI too.
My teams used this new AI for a range of novel applications, including training smart cameras on drones to find people lost in the wilderness, detecting when car doors were being opened in front of cyclists, and counting the number of desks in an office that were in use during the day. Additionally, we explored the ethical implications of these new AI capabilities and how an organisation can use them responsibly.
Now it seems AI has been redefined all over again, and generative AI is what people mean when I talk to them about AI. Which is a lot at the moment. Almost every professional conversation seems to turn to AI at some point. It’s a very exciting time, and there seem to be revolutionary announcements every month concerning generative AI.
Of course, this hasn’t escaped the notice of Boards and CEOs, who are asking their people to come up with an AI strategy for their organisations. Key suppliers are also putting pressure on these organisations to adopt their AI-enabled products and services, often with additional fees involved, and no CEO wants to fall behind competitors who are presumably “using AI” in everything.
It reminds me of the quip about teenagers and sex – and there are similar incentives here to talk about doing it, even if you’re not sure about it, and in fact aren’t doing it at all.
Actually, most organisations don’t need to get too worked up about it. It will be an evolutionary technology adoption for them rather than a revolutionary one, assuming they are already on the AI journey (AI meaning data science and deep learning).
This post is an outline of what a simple AI strategy can be for many organisations. Essentially, if an organisation is (i) not building software itself that appears in the user interface of its products and services, and (ii) has already adopted best practices for the previous generation of AI, it can likely keep things simple.
What’s new?
Generative AI can be considered an application of deep learning where new content is created, specifically audio, imagery or text that is similar to what a human would create. The recent AI boom has been brought about through a technology called a transformer architecture – the T in GPT stands for Transformer. Even before the excitement around OpenAI’s DALL-E 2 or ChatGPT services, you may have unknowingly used this technology in Google’s Translate service or Grammarly’s authoring tool.
While previous AI technology has been used in enterprises in decision-making tools, Gen AI has obvious application in creative tools. In a real way, the latest form of AI simply brings human-level capable AI-enabled features to a new set of enterprise tools. The insight is that you can treat this latest AI revolution as an update in enterprise tools. It may even be less disruptive than the time when enterprise tools moved to the cloud to be provided under SaaS (Software as a Service) arrangements.
When I say creative tools and decision-making tools, here’s what I mean:
- Creative tools are not just tools used by “creatives” but any tool used by people to create something new for the organisation. They include software development tools, word processing tools, graphical design tools and inter-personal messaging tools.
- Decision-making tools are any tool that provides data and insights that aid in making a business decision, such as to find the correct policy document, highlight the best applicants for a role, or report on monthly financial figures. The enterprise document repository, timesheeting system, or monthly dashboard are decision-making tools.
There are also some tools that are a mix of these two, for example Microsoft Excel allows people to create new financial models for their organisation that aid in making business decisions. That said, this hybrid category can be practically treated as a subset of decision-making tools.
In this discussion, I am assuming that the organisation in question has already done the usual things for the previous generation of AI. For example,
- evolved the data warehouse into a data lake that is able to store both structured and unstructured data ingested from operational and customer-facing platforms,
- established data governance processes and data management/ownership policies consistent with a relevant responsible AI framework (e.g. the Australian government ethical AI framework), and
- provided training around privacy, data sovereignty, and cyber security practices to people who handle business and customer data, or develop and test applications using it.
It is likely that the responsibility for doing all those things was with a part of the organisation that also had responsibility for the decision-making tools used in the enterprise, namely the IT team. Understandably, the IT team is probably where the Board and CEO are looking to get the AI strategy from.
Before we continue, let’s be clear about what AI will bring to creative tools. The following table provides examples of AI-enabled features used in different types of enterprise tools:
| Type of tool | Example AI-enabled feature |
|---|---|
| Decision-making tool | Forecasting |
| Classification | |
| Recommendation | |
| Search | |
| Anomaly detection | |
| Clustering | |
| Creative tool | Summarisation |
| Translation | |
| Transcription | |
| Composition |
What a particular feature does in a particular tool will be very tool-dependent. For example, in Adobe Suite, a composition feature might in-fill a region of an image to seamlessly replace the part that has been removed, while in Microsoft Powerpoint, a composition feature might provide initial layout of text and images on a slide. However, the high-level user experience is the same in both cases: the user provides a text prompt and receives new content in response.
Some decision-making tools are gaining a creative layer on top of their existing AI-enabled features, such as summarisation being added to search tools to save the user having to click on results, or language translation being added to recommendations to supported a wider user base. However, existing AI policies and procedures that have focused on decision-making tools will have likely picked-up these cases, and those tools that are a hybrid of decision-making and creative tools are well.
So what?
Organisations that produce creative tools will already have had to include Gen AI features in their products, driven by the customer/market demand for these and competitive pressures. These organisations will have had to skill-up in Gen AI already and have a good handle on the technologies and issues. This post is not for them.
Additionally, organisations that develop customer-facing software outside of creative tools will be considering how and whether AI-enhanced features like summarisation and translation could be incorporated in their user interfaces. The speed of innovation in this area is daunting. A year ago Meta’s foundation Gen AI model called Llama was leaked, initiating widespread development of such models in the research and startup communities, and now alternative models are beating OpenAI’s own models on public leaderboards (see here or here). There also also many complex factors to be considered. At the very least, such organisations should be performing upskilling in this area for their people and have a Gen AI sandpit environment for experiments. Given the speed of change in the marketplace, most organisations will need extremely quick ROI on any Gen AI projects or risk a waste of their investments. Due to all of that, this post is not for these organisations either.
If an organisation doesn’t build software that appears in the user-interface of its products and services, and given that Gen AI created text, imagery or audio will appear in user-interfaces, such organisations will be consumers of Gen AI rather than producers of it. I contend that the most common way for such organisations to consume Gen AI will be via tools that embed Gen AI, and hence avoid the costs and risks of building their own custom tools. Hence Gen AI technology adoption becomes a question of tool adoption and migration, and if an organisation has already tackled the question of AI before, it will have covered decision-making tools, leaving only creative tools to be dealt with in its plans.
Focusing on AI-enabled creative tools, these will have a number of common issues that an organisation will need to consider as part of adopting them:
- Copyright. New content is covered by copyright laws, which are similar around the world, but are not identical, and AI tends to play in the parts that are not globally consistent or well-defined, such as the concept of “fair use“. The data that has been used to train Gen AI models might turn out to have legal issues in some countries, impacting the use or cost of a model. The output of a Gen AI model may not be copyrightable, and hence others will be able to copy it without breaching copyright. This may limit how such AI models are able to be used in an organisation.
- New users. While the IT team has had its arm around the enterprise data warehouse and data lake, when it comes to creative tools, the IT team may not have been so involved, and adopted more of a light touch approach. The users of creative tools may not have received the previous round of data training, and may not be enrolled in data access systems intended to comply with data sovereignty controls, etc. From the point of view of AI, a Word document or corporate video is just as much “data” as the feed from the CRM.
- Data leakage. The latest Gen AI features in creative tools currently do not typically run on the desktop or on a smartphone, but are a SaaS feature that involves sending content to the cloud, and possibly off-shore. This is in many ways a standard SaaS issue rather than something new, but the nature of AI models is that they improve through training on data, so many tool providers seek to use what might be confidential content in the training of their models in order to continue to stay competitive. For example, Zoom modified their terms of service so that if a meeting host opts-in, the other participants in a meeting may have their meeting summary data used for training. Organisations are having to implement measures to manage this risk, such as Samsung choosing to restrict the use of ChatGPT after employees leaked confidential data to the tool last year.
- Misrepresentation. AI-enhanced creative tools might be used to produce content that others mistakenly think was produced by people or was otherwise authentic content. In the worst case, “deepfakes” may be created of an organisation’s public figures in order to dupe investors, customers or employees into bad actions. Scammers used this technique to trick a Hong Kong employee into transferring HK$200M. Still, a simpler case is where a chatbot on the Air Canada website made a mistake in summarising a company policy, a customer relied on this, and Air Canada was liable. Some organisations are taking care to carefully distinguish AI content from human-created content to help limit risks here.
Despite these issues, there is some optimism that AI-enhanced creative tools will bring a productivity boost to their users. The finger-in-the-air number is typically something like a 20% improvement. Microsoft’s recent New Future of Work Report (always very interesting!) includes some findings that Microsoft hopes will lead to uptake of their new AI-enhanced tools called Copilot:
- Copilot reduces the effort required. Effects on quality are mostly neutral.
- New or low-skilled workers benefit the most.
- As people get better at communicating with [AI tools], they are getting better results.
The Wall Street Journal covered some scepticism about the benefits of AI, highlighting that errors in AI output take additional effort to catch and correct, and there was a 20% drop in usage of some AI tools after the initial month of enthusiasm. This indicates that early adopters need to go into this with their eyes open.
Now what?
For organisations not building software that surfaces in the user interface of its products and services, the main impact of Gen AI will be on how and when to migrate to AI-enabled creative tools that their employees will use. Since the previous AI boom will have resulted in foundational AI procedures and governance in the organisation that can be reused for Gen AI, a simple AI strategy is to treat this shift to a new toolset as a change management exercise.
Further, instead of treating the migration of each tool as a separate exercise, it is worth managing this in a single program. There is a lot that will be common around managing the issues and conducting the training, so it will be more efficient to do it together.
An organisation will typically have a standard or preferred change management approach or blueprint for implementing technology change. This can be re-used for driving the migration to AI-enabled creative tools. No need to reinvent the wheel. (As an example, see the Bonus Content below for how the Kotter 8-step process might be tailored for this.) Note that the existing data governance processes will need to be leveraged in this process exercise. Additionally, the IT team will be fundamental in driving good use of Gen AI adoption.
In tackling the issues mentioned above, here are some questions to help work through the right path:
- Copyright. Which legal jurisdictions does the organisation and its creative tool suppliers operate in, and how do copyright laws vary (particularly the concept of “fair use”)? How important is having copyright over the output of creative tools, and are there other IP protection measures (e.g. trademarks) that mitigate any risks?
- New users. What degree is an organisation’s creative work done within the organisation, or done using external agencies/firms? How well do the legal agreements covering this work (whether employment or agency agreements) anticipate the issues of Gen AI? Is there consistency between how creative tools and decision-making tools are treated and managed in the organisation?
- Data leakage. Do people in the organisation understand how prompts and images given to Gen AI tools can leak out? What regulatory data compliance rules apply to data shared with or generated by these tools? How well do either “fine tuning” or “RAG” approaches to AI model customisation sit within an organisation’s risk appetite?
- Misrepresentation. How well do the official communications channels used by the organisation provide authentication? Are human and AI generated watermarking standards in use, e.g. Adobe Content Credentials or IPTC Photo Metadata standards? To what extent are misrepresentations of people at the organisation tracked and detected on social media? Which Gen AI web-scraping tools are blocked from ingesting the organisation’s public content?
You don’t need to over-bake it. For many organisations, the adoption of Gen AI will be through its enterprise tools, so it can be treated like a migration exercise. Just keep it simple.
(Thanks to Sami Makelainen, who provided comments on an earlier version of this post.)
Bonus content – Kotter 8-step process example
Here’s an example of how you might include activities within the Kotter 8-step change management process to help an organisation migrate to AI-enabled creative tools:
- Create a sense of urgency. Identify how the use of Gen AI tools links to the organisational strategy (improve staff experience, greater productivity, etc.) and an answer to “why now” (CEO directive, culture of leadership, existing strategic program, etc.).
- Build a guiding coalition. Ensure senior stakeholders have bought in to this rationale, with a single influential stakeholder willing to represent the activity. Ensure parts of the organisation outside of IT are represented, such as vendor management, legal, and parts of the organisation that use creative tools, e.g. anyone with “manager” in their title. Ensure the working group is suitable trained about Generative AI technology and its emerging issues, such as those outlined above.
- Form a strategic vision. With the stakeholder group, develop a view of how the organisation will be different once it has migrated to new AI-enabled tools, e.g. include use cases. This should be tangible and time-bounded, so should ideally be informed by previous tool migration exercises.
- Enlist a volunteer army. Leverage internal organisational communications tools to promote the vision, build a cross-organisation community of supporters. People are generally pretty excited about this new application of AI. The stakeholders and community can together help expand the community so it is truly cross-organisational. Task them to identify the creative tools that are used across the organisation (including “free” tools), which ones already have AI-enabled features, what types of data are consumed and generated by these tools, which suppliers provide them, and where the data is processed. Identify simple metrics that would highlight if the features of these tools successfully bring the expected organisational benefits.
- Enable action by removing barriers. Ensure the community gets training about the issues relating to AI-enabled creative tools. Leverage the community to consider the risks of different uses of these tools in their different parts of the organisation, determine what constraints should be applied around the use of these tools, e.g. when can confidential information be shared with the tool. If the constraints are onerous, identify if alternative tools exist that could have fewer constraints.
- Generate short-term wins. Focus on one or two tools, prioritising those with the most benefit and easiest to migrate. It may be that it is easiest to start with something like GitHub Copilot and some software engineering teams, or maybe it will be easiest to use something like Microsoft 365 Copilot and some people with “manager” in their titles. Gain agreement to migrate these initial tools and learn from them. Ensure the users of these tools are trained to use the tools under the constraints, and specifically on writing good prompts. People who are already using AI-enhanced tools in the community may be a good source of training information.
- Sustain acceleration. Track the metrics to see where the migration to AI-enhanced tools has brought the expected benefit. Use the learnings to build a business case for migrating more tools and leverage the stakeholders to drive the wider adoption of AI-enabled creative tools.
- Institute change. Not everything will have gone smoothly. Update policies and procurement practices to accommodate learnings. Provide organisation-wide training on Generative AI technology, and use the community, stakeholders and metric data to bring the rest of the organisation up to speed on the new tools.