The Accelerating Pace of AI
Copy this iframe to your website:
1 | |
Description
This MicroSim places 23 milestones in the history of artificial intelligence, from the 1956 Dartmouth Conference to 2025, on one time axis. Each milestone is a stem with a circle on top. The stem color shows the era, and the legend counts the events in each era: 5 before 2000, 5 in 2000-2014, 6 in 2015-2022 and 7 from 2023 on. Hover over a circle to read what happened.
The Time-axis scale slider changes how years are mapped to horizontal position. The position of a year along the axis is \(\left(\frac{\text{year} - 1956}{69}\right)^p\), where the exponent is \(p = 1 + 2.5 \times \text{slider}/100\). At the left end \(p = 1.00\) and the axis is linear. At the right end \(p = 3.50\), 1956 and 2025 stay at the ends, and every year in between moves left, so recent years get more room. Stem height is only a layout device that grows with time. It is not a measurement.
AI methods for signal processing are covered in Chapter 15: Signal Processing Applications and AI Integration.
Lesson Plan
Learning Objectives
After using this MicroSim, students will be able to:
- Identify the era of a milestone from the color of its stem and the legend
- Compare how many milestones the plotted list contains in each era
- Predict how a larger exponent \(p\) moves each milestone along the time axis
- Evaluate whether a hand-picked event list is enough evidence that AI progress is accelerating
Target Audience
College undergraduates in a signal processing or introductory AI course. Allow 10-15 minutes.
Prerequisites
- Reading a timeline
- Powers and exponents, for example that \(0.5^2 = 0.25\)
Activities
Activity 1: Linear view (3 min). Leave the slider at 0 (\(p = 1.00\)). Read the legend: 5, 5, 6 and 7 events. The 7 red events share only three stem positions (2023, 2024 and 2025), all in the last 2 of the 69 years, about 3% of the axis.
Activity 2: Expand recent years (4 min). Drag the slider to 100 (\(p = 3.50\)). The 5 events before 2000 move into the left 16% of the axis, and the red events spread out: 2023 now sits at 90% of the axis instead of 97%.
Activity 3: Predict, then check (3 min). At \(p = 1.00\) AlexNet (2012) sits at 81% of the axis. Predict whether it moves left or right when you drag the slider to 100, then check. It moves left, to 48% of the axis.
Activity 4: Read the descriptions (3 min). Hover over Transformer Architecture (2017), BERT (2018), GPT-3 (2020) and ChatGPT (2022). Note the gaps between them are 1, 2 and 2 years. Read each description and find the wording that links BERT to Transformers.
Assessment
- How many events does the legend count for 2023 and later, and how many separate stem positions do they use at \(p = 1.00\)? (7 events on three stem positions, one each for 2023, 2024 and 2025.)
- The five events before 2000 are 1956, 1966, 1980, 1990 and 1997. The six events in 2015-2022 are 2016, 2017, 2018, 2020, 2021 and 2022. Which group has the smaller average gap between neighbors? (2015-2022: the gaps are 1, 1, 2, 1 and 1 years, an average of 1.2 years, versus 10, 14, 10 and 7 years, an average of 10.25 years, before 2000.)
- At the right end of the slider, what is \(p\) and where does Deep Blue (1997) sit? (\(p = 3.50\) and \((41/69)^{3.5} \approx 0.16\), so Deep Blue is at about 16% of the axis.)
- Does this MicroSim show that AI progress is accelerating? (Not by itself. It plots a hand-picked list of 23 milestones, so it shows that the list is denser in recent years. Counting events does not measure capability.)
Common Misconceptions
"A taller stem means a more important or more capable system." Stem height grows with the year and is nudged up so neighboring names do not collide. Only the year, the name and the description are data.
"The list shows how fast AI improved." The list is a selection, and recent products are easier to list than older research. Compare the slider at 0 and at 100: the clustering depends on the axis, and the count of events is not a measure of performance.
References
- Timeline of artificial intelligence - Wikipedia - a longer, sourced list of AI milestones to compare with this one
- History of artificial intelligence - Wikipedia - narrative history of the field and its funding cycles
- Dartmouth workshop - Wikipedia - the 1956 event that starts the timeline
- Transformer (deep learning architecture) - Wikipedia - the 2017 architecture behind BERT and later language models
- Exponential growth - Wikipedia - what a measured exponential trend looks like, for contrast with a count of events
Development Notes
Version 2.0.0 (2026-10-14). Ported to p5.js 2.3.2 and standardized. The slider's response was discontinuous in the earlier sketch: it jumped from a near-cubic axis back to a linear one at the middle of the slider, and its right end was labeled "Logarithmic" although the mapping is a power law that expands recent years. It is now one monotone mapping, \(p = 1 + 2.5 \times \text{slider}/100\), and the sketch starts at the linear end (the same first view as before). Name labels now avoid each other, hover detection matches where the circles are drawn (it was off by 10 px horizontally and 50 px vertically), and the years of DeepSeek R1 and the Y-Combinator coding survey were corrected to 2025.