How Long Before Half of People Use Generative AI?

One key question observers have about generative artificial intelligence is how long it might take before usage reaches “mass adoption.” That obviously requires creating benchmarks, and we often use “years” and “percentage of population” as metrics.

So we might measure time from the founding of a firm and any of its products reaching 10-percent usage by consumers, and then perhaps 50-percent usage, on the way to creating graphs showing either growth rates or adoption.

“Adoption” then is tricky to the extent that we have to define what “usage” actually means. As always, is it daily or monthly users? Are they regular users? And if so, what numbers do we use for those thresholds? Also, in many cases we have to make assumptions about the percentage of such users in specific countries, since availability or usage often is not the same everywhere. 

For consumer products, many researchers seem to agree that a 10-percent level of adoption encompasses the early adopters and tends to be the trigger for widening adoption beyond the “tech enthusiast” stage. Most of us would be comfortable with the notion that 50-percent adoption is a reasonable enough definition of “mass adoption,” or “commonly and regularly used,” even if it is not true that “everybody uses it.”

So, using company founding; 10-percent and 50-percent adoption levels, many popular consumer products and services took two to eight years to reach 10-percent adoption, and perhaps seven to 26 years to reach 50-percent use.

Consumer Adoption of New Technology
Product	Company Founded	10% Use (Years)	50% Use (Years)
Google Search	1998	2001 (3)	2008 (10)
Amazon E-commerce	1994	2001-2002 (7-8)	2020 (26)
Facebook	2004	2007 (3)	2013 (9)
Uber	2009	2015 (6)	45% by 2023
Airbnb	2008	2015 (7)	35-40% by 2023
YouTube	2005	2007 (2)	2012 (7)

Though it is unclear whether generative AI is more like YouTube or more like Amazon e-commerce, and assuming we date the start of the product to 2023 (ChatGPT), 

The caveats are that we do not yet have a good feel for how to define “regular use” or adoption. Much of the early ChatGPT use might have been “sampling,” where people used it just to get a feel for it, but might not have followed up with regular use. That caveat notwithstanding, it does seem that generative AI is being adopted faster than most earlier popular internet services. 

In fact, combining usage of several popular or early services, GenAI is being adopted faster than any of the popular internet consumer apps. 

Service	Launch Date	Estimated 10% Date	Years to 10%
ChatGPT	Nov 2022	Feb 2023	0.25 (3 months)
Google Gemini	Dec 2023	Jun 2025	1.5
Claude	Mar 2023	Mar 2026	3
Perplexity	Jun 2022	Dec 2024	2.5
Generative AI (all)	Nov 2022	Nov 2023	1

Again, the caveat is that it is unclear what percentage of those estimated users are “regular and habitual” users, and what percentage are samplers who do not use the tools regularly.

For that matter, even the estimates of internet consumer app usage might be on the high side, as well. 

Google search by 2000 handled 100 million searches per day. Assuming many were U.S.-based and accounting for repeat users, it likely hit 28 million monthly users (10 percent of U.S.) by 2001 (three years). 

Amazon e-commerce had 20 million customers by 2000, mostly in the United States, and probably reached 28 million users  (10 percent) by 2001 or 2002.

Facebook reached 12 million users by 2006, mostly in the United States and probably reached the 10-percent level by late 2007 after opening to the public.

Uber likely hit 33 million users 10 percent) by 2015 (six years), assuming U.S.-centric early growth.

Airbnb probably hit 33 million users  (10 percent) by 2015. 

YouTube grew to the 10-percent level in about two years.

The point is that, even if generative is as popular as we think, and might by some measures already reached the 10-percent level of use, it still could take up to seven years (if it follows the YouTube growth curve) to reach 50-percent usage by U.S. users. Professional and business users, as well as students, are likely to drive much of the early usage.

Building Agentic AI is Akin to Building a Car

For us essentially non-technical types, it often is hard to understand what all is involved in creating an agentic artificial intelligence system that acts essentially autonomously on a user’s behalf. It isn’t so much the coding details but understanding what the system or machine has to do. 

Compare that to understanding the capabilities an automobile must possess to function. Autos need systems for power, steering, stopping, fuel supply, navigation, sensors, transmissions and dashboards. 

Agentic AI systems likewise must possess any number of capabilities supplied by sub-systems, modules or platforms that supply those features. 

Automobile Capability	Agentic AI Capability	Explanation
Engine (Power Source)	Computational Efficiency and Resource Management	The AI must have efficient computing power to process information and execute actions.
Steering System	Goal-Oriented Planning and Decision-Making	The AI needs the ability to set objectives, make decisions, and adjust its course dynamically.
Brakes,  Safety Mechanisms	Ethical Constraints and Self-Regulation	Ensures the AI does not cause harm, stays within boundaries, and avoids unintended consequences.
Fuel or Battery	Access to Data and Knowledge Base	Just as fuel powers a car, AI needs continuous data input to function effectively.
Navigation, GPS	Situational Awareness and Context Understanding	The AI must be aware of its environment, interpret contexts, and adapt accordingly.
Sensors (Cameras, Lidar.)	Perception and Multimodal Understanding	AI should process inputs from multiple sources (text, images, audio, real-world sensors).
Transmission System	Task Execution and Action Mechanisms	The AI must translate decisions into meaningful actions efficiently.
Dashboard and Controls	User Interaction and Explainability	The AI should communicate with humans, explain decisions, and receive feedback.
Maintenance and Diagnostics	Self-Improvement and Learning	The AI must refine its models over time, detect errors, and optimize performance.
Emergency Handling	Failure Recovery and Adaptability	AI needs contingency mechanisms to handle unexpected issues and correct itself.

An agentic AI system might not be as complicated an endeavor as Artificial General Intelligence (AGI), but it still is plenty complicated, involving a platform that integrates any number of capabilities including reasoning and problem solving; learning and adaptation; communication and understanding; perception; decision-making and planning; memory and knowledge management; goal setting; self-monitoring and reflection; human interaction; creativity; autonomy; security and adaptability. 

As you would guess, each of these capabilities is embodied in different platforms or systems. 

Reasoning and Problem-Solving:

• Deductive Reasoning: Prolog, Expert Systems like CLIPS

•  Inductive Reasoning: Decision Trees, Bayesian Networks (e.g., Bayes Net Toolbox for MATLAB).

• Abductive Reasoning: Systems like Sherlock (a framework for abductive reasoning).

• Problem Decomposition: Planning systems like PDDL (Planning Domain Definition Language) used by planners like Fast Downward or LAMA

Learning and Adaptation:

• Supervised Learning: TensorFlow, PyTorch for training models like CNNs, RNNs.

• Unsupervised Learning: Autoencoders, K-means clustering implemented in scikit-learn.

• Reinforcement Learning: OpenAI's Gym, DeepMind's MuJoCo, or libraries like Stable Baselines3

Communication and Understanding:

• Large Language Models (LLMs): Hugging Face's Transformers with models like BERT, GPT-3, or custom implementations like those from xAI.

• Transformers: Libraries like Hugging Face Transformers or Microsoft's DeepSpeed

Perception:

• Image Recognition: OpenCV for basic operations, TensorFlow or PyTorch for CNNs like ResNet, YOLO for real-time object detection.

• Speech Recognition: Kaldi, Mozilla's DeepSpeech, or cloud APIs like Google Cloud Speech-to-Text.

• Sensor Integration: ROS (Robot Operating System) for robotics, or custom IoT frameworks for sensor data integration

Decision Making and Planning:

• Rule-Based Systems: Drools, Jess for Java, or custom Python rule engines.

• Neural Networks: Keras, TensorFlow, PyTorch for implementing various neural network architectures

Memory and Knowledge Management:

• Short and Long-Term Memory: Implementations of LSTM or GRU in neural network libraries for memory, or dedicated systems like MemN2N (Memory Networks).

• Knowledge Graphs: Neo4j for graph databases, or RDF systems like Apache Jena for semantic web applications.

Also required: 

• Goal Setting and Management: Custom implementations, or frameworks like BDI (Belief-Desire-Intention) agent architectures.

• Ethical Decision Making: Not standardized but could involve rule engines with ethical guidelines or AI fairness toolkits like Fairlearn.

• Self-Monitoring and Reflection: Could involve meta-learning frameworks or custom solutions using reinforcement learning for self-improvement.

• Interaction with Humans and Other Systems: Dialog systems like Rasa or Microsoft Bot Framework, or APIs for system integration.

• Creativity and Innovation: Generative Adversarial Networks (GANs) using TensorFlow or PyTorch, or systems like DALL-E for image generation.

• Autonomy in Execution: Custom agents using frameworks like JADE (Java Agent DEvelopment Framework) or integrating with IoT platforms for physical actions.

• Security and Privacy: Cryptographic libraries like OpenSSL, or frameworks ensuring differential privacy like TensorFlow Privacy.

• Adaptability to New Environments or Tasks: Meta-learning approaches like MAML (Model-Agnostic Meta-Learning) or transfer learning capabilities in deep learning frameworks.

• Reasoning and Problem-Solving:

• Deductive Reasoning: For logical conclusions from known premises.

• Inductive Reasoning: To infer general rules from specific instances.

• Abductive Reasoning: For forming hypotheses from incomplete data.

• Problem Decomposition: Ability to break down complex problems into manageable parts.

The point is that for all the legitimate attention now paid to large language models (generative AI), that field is but one among many that would have to be assembled and orchestrated to create agentic AI.