From Tee to Green: Navigating AI's Hype Cycle Fairway
Humanity has grown through its hardships, with each deep wound making it tougher for the next generations to heal. No matter how many technological breakthroughs we make, we're still failing to address conflicts, prolonging the agony and distress that people endure. Our species has come a long way, but always at a steep price — and it doesn't look like that's going to change anytime soon. As I begin to write, my usual excitement for the latest technological breakthroughs is dampened by a sense of sorrow, influenced by current events worldwide. I hope that we, as a human race, can find even a sliver of compromise in our rewriting of history to pave a path toward future healing.
Last week, another issue took the spotlight. We gave birth to Alien Intelligence this year which is infused with humanity’s collective knowledge up to this point. This has stirred a ‘fear of the unknown’ among us, with concerns it might one day develop a super-intelligence that surpasses our own species. Global consortium concluded regulatory oversight is a must now to ensure the right level of guardrails are put in place with AI as it co-create, co-exist with Human intelligence. It's interesting to watch the two sides to this: on one hand, there's the need for regulation, but on the other hand, too many rules might slow down innovation. I'm going to leave this argument for the future to figure out. It was interesting to see AI leaders release powerful AI models within few days after signing the Safety act to regulate!
As 2023 comes to a close, I find myself pondering: Are the initial stages behind us? Have we played through the early, perhaps simpler holes? Do we understand the lay of the course well enough, and are we mentally prepared to tackle the more challenging back nine?"
As I consider the future, I realize that definitive answers are elusive. If I do come across a promising insight, I approach it with caution, knowing how unpredictable the future can be. In search of some guidance, I turned to frameworks that can help navigate through established routes.
According to the Gartner Hype Cycle, General AI is currently in the "Trough of Disillusionment" stage, and I find this analysis accurate. We've come to a point where, although the innovation engine is humming,, the actual application of General AI for widespread productivity is under careful examination. Experiments underway to build an acceptable range of trustworthiness, and most importantly proof of value creation at reasonable ROI is calculated. On the technical side, focus is on implementing strong security measures, developing models tailored to specific business needs, and employing techniques like Retrieval-Augmented Generation (RAG) and Prompt Engineering to assure the reliability of General AI systems. What I'm watching closely is the "Time to Value" gap – the time it takes to move from high expectations (around mid-2023) to a level of steady productivity. The question is, will this transition span years, or will it stretch into decades? What I saw today from Open AI Dev Day, we are sure to have substantial productivity gains in a few years!
Before exploring the stages of the Hype Cycle, it's important to acknowledge AI's potential to equalize opportunities for everyone, regardless of gender, race, or economic status. From my own hands on experience and observations of how people utilize these current sets of AI tool kits, the impact of AI assistants in enhancing our writing, creativity, coding and ideation is extraordinary. This great research study amplifies it. Worth a read!
Innovation Trigger Phase:
In November 2022, the Innovation Triggered millions overnight to subscribe to ChatGPT. This marked a turning point offering a conversational interface that is almost indistinguishable from human interaction. We accepted this phase rather quickly, experimented, adopted it in our daily needs of creative work and predicted in our own view of the situation as a technological pivotal moment in history. Pundits called this the Digital Renaissance, Economists predicted major disruption to the human labor market immediately, especially the knowledge worker categories and long term displacement. AGI predictors revised the target dates quite substantially based on these innovation triggers.
Peak of Inflated Expectations Phase:
GenAI technologies reached a Peak of Inflated Expectations in Mid 2023. Investment and strategies re-defined everywhere in urgency, M&A and alliances emerged demonstrating incumbent dominance, every product innovations exhibited unparalleled creativity setting new levels of expectations. My personal monthly subscriptions to consume GenAI related technology services pivoted dramatically ;) [ 20$ ChatGPT, 10$ MidJourney, 5$ ElevenLabs, 8$ DreamStudio, 6$ D-id.com, 18$ OpenAPI ]
Multi-billion $ investments in AI startups soared. Pitchbook data indicated the value of funding for AI companies climbed 27% globally in the 3rd quarter compared to last year. Amazon stake with Anthropic 1.25B$ investment and Meta’s constant drive to deliver innovative open source stack for broader adoption, Elon Musk’s xAI formation and many more, all to gain a portion of the inflated expectations from GenAI technology.
Every extension of the foundational models recently delivered enabled a new set of use cases and new set of predictions to disrupt the labor force. Co-Pilots, PersonalAI assistants, AI Tutors, Humanoid robots and many more tool kits demonstrating what would come next. Now, the race is to be the best multi-modal foundational models (vision, audio, video etc) and there is a heightened level of urgency to deliver this and take the throne of “best model yet” beating GPT-4v, GPT4-Turbo now!
Trough of Disillusionment Phase:
While the promises of productivity gains are unimaginable with GenAI, the reality of when we hit that plateau of productivity curve is yet to be seen. We now have entered the Trough of disillusionment phase. Few key data points that I feel would resonate this state.
Unable to find a broader successful sticky use case leveraging GPT4 or any other models. We don’t have an Uber app moment yet on the GenAI LLM platforms. We do have lots of applications getting enabled with GenAI as co-pilots but we have not found an AI-First-Use-case. The chances of this happening in consumer space is much higher as wider proliferation enters our digital life.
Enterprise readiness: No P&L demonstrated an impact with GenAI either to the bottom line or top line. Nor do we see margin impacts due to additional R&D expenses for model training or finetuning. While GenAI Infrastructure Cloud providers are spending in billions, the impact for end enterprise P&L is not yet seen.
Trust score: Successful stories of decision making on outcomes or complex automation of tasks via GenAI yet to be seen. Skepticism due to hallucinations is a real challenge to gain trust on the GenAI Applications.
Now, let’s look at what it would mean to transition from Trough of disillusion to Slope of enlightenment in the Gartner Hype cycle for GenAI. Being cautious on predictions, I am attempting to share my perspective on how genAI would transition in this phase of the hype cycle.
Transition to Slope of Enlightenment Phase:
Some key category of drivers:
Customer Centricity: Use Case & Value driven. Focus will shift to solving problems end to end with well defined sticky workflows and leverage multi-modal capabilities. We already started seeing this with OpenAI releasing CustomChatBots, an easy way to create your own GPT with your own data. We would see companies building custom LLM for a specific vertical and offering that as a service. Example:https://www.harvey.ai/blog Generative User Interfaces: We are very much accustomed to the current conversational interface (like ChatGPT) which spits responses. We are yet to get into Generative UI for everything we do digitally. Perplexity, Inflection AI are really leading the pack on this in my view. These types of interfaces will lead to broader adoption of complex workflows that lead to full automation of tasks, driving value, and rising productivity, all with Natural language.Personal AI Assistants: Always on tutor, mentor, coach, companion will be the future. Each one of us will have our own AI assistant trained on our data and tunable to our needs. Like Inflection AI, there will be many more that will focus on this consumerization of GenAI.
Trust, Trust, Trust: Overcoming hallucination risks With a wide array of research communities exploring, experimenting and defining possible solutions to overcome certain limitations of current foundational models, there will be a path to gain more trust. One of the approach which seems to gain momentum is RAG ( Retrieval Augmented Generation).
Emerging reasoning techniques like Chain-of-thought, tree of thought enable model capabilities to perform complex reasoning tasks aligning to customer expectations. My go to reference is prompt engineering techniques captured here.
Constitutional AI: Anthropic are pioneering this idea of Aligning an AI's behavior with a "constitution" defined by human principles - things like avoiding harm, respecting preferences, and providing true information. This constitution shapes how the AI acts.
3. Enterprise Readiness: To reach the level of maturity where Enterprise productivity is uplifted, there are quite a number of innovations to be driven. While many are set in motion with abundant investments, consolidations and alliances, I am sure there is more to be done.
Cost & Accessibility to train models: We are seeing emergence of AI First Infrastructure companies like Coreweave, Lambda Labs etc which provide GPU farms at a reasonable cost with pay as you grow models. Along with these, incumbent major cloud providers gear up their AI Infrastructure services with easy to train, build, deploy LLM in a secure enterprise infrastructure domain. Plus these large Cloud providers will start cobbling startups to drive their innovation engine at a much faster pace and eventually drive lower cost to Enterprises. Open AI GPT4 Turbo with 128K context length is 3xCheaper than GPT4!
Enterprise focus Services - Bing Enterprise, Google Vertex AI Enterprise search, Open AI Enterprise subscriptions for Teams (rumor) or Open AI Apps Marketplace (rumor). Introduction of various co-pilots in the current set of enterprise productivity tools like Google suites, Adobe, Salesforce, Servicenow, GitHub, O365, SAP etc is a natural progression towards AI first use cases for future. Adoption of these into the Enterprise ecosystem will be possible when these co-pilots prove its safe, secure and data privacy is maintained while providing Enterprise specific outcomes.
Domain specific - purpose led models : This will be the key differentiator for Enterprises to evolve and gain more trust, confidence in GenAI pertaining to their data sets, their automation needs, and their productivity improvement plans. Building these domain specific GPT’s are made easy as seen in Open AI Dev Day.
Level 0 to Level 5 Autonomy with AI Agents: As the enterprise journey progresses, GenAI Agents would be the next wave of innovation which would dramatically improve productivity in Enterprise settings. For example: Microsoft AutoGen framework competing with LangChain framework for LLM agents. This framework enables simpler interfaces to external LLM agents/databases to perform a complex set of tasks.
In summary, We have to admit that AI we knew from the past was linked to repetitive tasks, the type of work that could be substituted by human’s left brain tasks. Now, GenAI is participating in the right side of the brain tasks (co-creation, idea generation, logical thinking via multi-modal support). I came across this great analogy “Gen AI will be the bulldozer for the mind” Eventually, GenAI & its future extensions with GenAIAgents (caution:me coining this term!) plus these traversing the adoption curves in becoming general purpose technology, the impact on human productivity & society as a whole is unknown. When we see Enterprises or Consumers articulating their P&L dramatically change due to productivity gains with GenAI and its sustainability for the long term, we are all set to transition from the Slope of Enlightenment to Plateau of productivity.
We got a glimpse of the art of possibility at Open AI Dev Day!
Exciting times indeed, we, the humans, will witness this!