If you could never forget anything, you would not become a genius; you would become someone whose past keeps playing whether you want it to or not. The closest real people, about 50 confirmed cases of highly superior autobiographical memory (HSAM), recall the day and events of almost every date for decades, yet score normally on many lab memory tests and form false memories like everyone else. And a brain that literally kept every input would need roughly 1 petabyte for a lifetime of vision alone, against an estimated capacity of 50 to 500 terabytes.
- About 50 people with HSAM have been identified by the UC Irvine team from several hundred candidates.
- HSAM participants recalled 85% on a 10-dates test versus 8% for controls, but matched controls on digit span and paired words.
- They still form false memories: equal or higher rates than controls in a 2013 PNAS study.
- 4.7 bits per synapse is the best measured estimate; scaled to a brain it gives roughly 50–500 TB.
- Ebbinghaus forgetting follows a power law: about 15% savings left after a year, never quite zero.
Forgetting feels like a bug. You forget where you kept the keys, the name of the man at the wedding, the formula for the exam. But neuroscientists increasingly treat forgetting as part of the design. Total recall would change three things you might not expect: how you decide, how you grieve and how you forgive.
Are there people who really never forget?
Yes, within limits. Highly superior autobiographical memory, also called hyperthymesia, was first described in 2006. People with it can be given a random date from decades ago and tell you the weekday, what they did and what was in the news. What they do not have is a photographic memory for everything.
In the late spring of 2000, neurobiologist James McGaugh at UC Irvine received an email from a 34-year-old woman named Jill Price. She said that since she was a girl she could name the weekday of any date from 1974 onward and what she had been doing, without diaries or calendars. McGaugh was sceptical and invited her to the lab on 24 June 2000. She correctly dated public events they named and named events for dates they gave. In 2003 she recalled the dates of 23 previous Easters with one error, along with what she did on each. She also struggled to match keys to locks and was poor at rote memorisation. The team's 2006 paper, which called her "AJ", described her remembering as "nonstop, uncontrollable, and automatic".
Scientific American: The Discovery of Super Memories, by James McGaugh and Aurora LePort (2014)That last line is the important one. Price did not experience her memory as a gift she could use. She experienced it as something happening to her.
What testing showed
A 2012 study by Aurora LePort and colleagues screened 115 people who claimed the ability and brought 11 into the lab. The numbers below come from that paper.
| Test | HSAM (n = 11) | Controls |
|---|---|---|
| Public Events Quiz (screening) | 56.5% | 12.9% |
| 10 Dates Quiz, day of week | 97% | much lower |
| 10 Dates Quiz, overall | 85% | 8.2% |
| Verifiable details in one week's recall | 33.9 | 11.3 |
| Accuracy of verifiable details | 10 of 11 at 100% | not compared |
| Forward digit span, paired words | no difference | no difference |
Eleven HSAM participants (aged 27–60) were compared with age- and sex-matched controls. They far outperformed controls on dates and personal events, but on standard lab tests such as digit span and verbal paired associates they were no better. Nine of 11 reported hoarding, a strong need for order or germ avoidance, and they scored higher on an obsessional inventory (8.09 vs 5.21), though none had a clinical OCD diagnosis. MRI found differences in nine brain regions, including the uncinate fascicle. Limit: tiny sample, and the brain differences may be a cause, an effect or neither.
Read the paper (free)
Does never forgetting mean your memories are accurate?
No. A brain that keeps more is not a brain that records more faithfully. In 2013 Lawrence Patihis, Elizabeth Loftus and the UC Irvine HSAM team tested 20 HSAM participants and 38 controls on three classic false-memory tasks.
- Word lists (DRM task): both groups "recognised" related words that were never shown.
- Misinformation: after seeing a photo slideshow and reading a misleading account, HSAM participants showed higher overall false memory than controls.
- Imaginary news footage: HSAM participants were as likely as controls to say they had seen footage of a plane crash that does not exist.
Twenty HSAM participants and 38 age- and sex-matched controls were tested with associative word lists, a misinformation paradigm and a nonexistent-footage question. HSAM participants were not immune on any task. The authors conclude that remembering is reconstructive even in people who remember abundantly and accurately. Limit: lab tasks with neutral material; personal memories might behave differently.
Read the paper (free)Even the people who remember everything are rebuilding it each time they look.
This links to a broader idea in neuroscience: the brain predicts, then corrects. A memory is regenerated from stored fragments plus present expectations. More on that model in predictive processing and maya.
How fast do normal people forget?
Fast at first, then very slowly, and never to zero. Hermann Ebbinghaus measured this on himself in 1880 with nonsense syllables, using the savings method: how much less time it takes to relearn a list than to learn it the first time. In 2015 Jaap Murre and Joeri Dros replicated the experiment, with one subject spending 70 hours learning and relearning lists.
See the data as a table
| Ebbinghaus (1885) (% saved) | Dros replication (2015) (% saved) | |
|---|---|---|
| 20 min | 58.2 | 47.2 |
| 1 hour | 44.2 | 37.3 |
| 9 hours | 35.8 | 27.6 |
| 1 day | 33.7 | 31.7 |
| 2 days | 27.8 | 23.0 |
| 6 days | 25.4 | 16.8 |
| 31 days | 21.1 | 4.1 |
Source: Murre & Dros, PLoS ONE 2015, Table 3 · Savings = share of the original learning time saved on relearning. Note the small bump at 1 day in both curves, which the authors link to sleep.
One subject learned and relearned lists of nonsense syllables after 20 minutes, 1 hour, 9 hours, 1, 2, 6 and 31 days, in about 70 hours of sessions. The curve closely matched Ebbinghaus's 1880s data, with an upward jump at the 24-hour point that suggests sleep helps consolidate. Limit: a single subject and meaningless material, which is forgotten faster than meaningful memories.
Read the paper (free)Solving the curve
Fit a power law to Ebbinghaus's seven points by least squares on log-log axes. The power law of forgetting says retention falls as a fixed power of elapsed time, so each doubling of delay costs the same fraction.
- Fit. Regressing \( \ln Q \) on \( \ln t \) (t in minutes) gives \( A = 0.80 \) and \( d = 0.126 \). This is my fit to the published table.
- One hour. \( Q(60) = 0.80 \times 60^{-0.126} = 0.48 \), close to the measured 0.44.
- One year. \( t = 525{,}600 \) min, so \( Q = 0.80 \times 525600^{-0.126} = 0.15 \).
Calculated in this post from the power-law fit above (A = 0.80, d = 0.126), which is my least-squares fit to the Ebbinghaus savings table in Murre and Dros (2015).
- Ten years. \( Q = 0.80 \times 5256000^{-0.126} = 0.11 \). Ten times the delay removes only a quarter of what was left.
Calculated in this post from the power-law fit above (A = 0.80, d = 0.126), which is my least-squares fit to the Ebbinghaus savings table in Murre and Dros (2015).
So normal memory already "never forgets" in a weak sense: a trace survives as savings. What the curve removes is easy access. A never-forget brain would be one where d is close to zero, and everything stays as accessible as it was an hour after it happened.
Would your brain run out of space?
If "never forget" means storing every input, yes. Here is the working, with every assumption visible.
Capacity side
- Neurons: 86.1 billion in the adult male brain (Azevedo et al., 2009).
- Synapses per neuron: my assumption, 1,000 to 10,000. As a cross-check, a cubic millimetre of human cortex held about 150 million synapses (Shapson-Coe et al., 2024); times a 1,200 cm³ brain that is 1.8 × 1014, inside the range.
- Bits per synapse: 4.7, from 26 distinguishable synapse sizes in rat hippocampus (Bartol et al., 2015).
Demand side
- Waking hours. Assume 16 hours a day.
- Bitrate. Assume vision alone, compressed like an HD video stream at 5 Mbit/s (my estimate; raw retinal input is far larger).
- Per day. \( 16 \times 3600 \times 5\times10^{6} = 2.9\times10^{11} \) bits, about 36 GB.
With these example inputs (16 waking hours a day and a 5 Mbit/s video bitrate, both my assumptions), this is simple arithmetic, not a measurement.
- Per lifetime of 80 years. \( 36\ \text{GB} \times 365 \times 80 \approx 1.05\ \text{PB} \).
That is 2 to 20 times the estimated capacity, before sound, smell, touch or thought, and assuming every synapse did nothing but storage. The ceiling here is a counting argument: Shannon's source-coding theorem says you cannot store a signal losslessly in fewer bits than its information content. So the brain must compress, and compression means throwing detail away.
Use this to check what you just read: change the assumptions and see whether a recorded life fits in your head.
Would a recorded life fit in your brain?
Drop the bitrate to 0.01 Mbit/s, roughly a diary of gist, and a lifetime takes about 2% of capacity. That is what HSAM brains appear to do: keep episodes and dates, not pixels.
With these example inputs (0.01 Mbit/s, 16 hours a day, 80 years, 1.8 × 1014 synapses), calculated in this post from the Bartol et al. (2015) figure of 4.7 bits per synapse: 1.7 × 1013 bits ÷ 8.5 × 1014 bits ≈ 2%.

Why is forgetting useful for decisions?
Forgetting keeps your model of the world up to date and stops you from overfitting to one-off events. That is the argument of a 2017 paper in Neuron by Blake Richards and Paul Frankland, who wrote that the goal of memory is not to carry information through time but to optimise decisions.
- Flexibility: forgetting reduces the weight of outdated information. The road that flooded in 2019 should not still feel like today's road.
- Generalisation: forgetting the details of individual events lets the brain keep the pattern. This is the same reason machine-learning models use regularisation.
- Speed: fewer competing traces mean less interference when you search.
Borges made the same point in fiction: his character Funes was troubled that a dog seen at 3:14 from the side should share a name with the same dog seen at 3:15 from the front. Perfect detail blocks abstraction. The limit on what you can hold in mind at once makes this worse, as covered in why working memory holds about four things.
Do HSAM brains forget on purpose?
They can. In a 2025 fMRI study, Valerio Santangelo and colleagues gave 12 HSAM participants and 30 controls words followed by "remember" or "forget" cues. Both groups forgot the "forget" words equally well, but HSAM brains showed extra frontoparietal activity when words first appeared and extra midline activity during forgetting, as if they had to work harder to let go.
What would never forgetting do to grief and grudges?
It would slow both down. Normal grief softens partly because the sharpness of specific memories fades while the meaning stays. People with HSAM describe old embarrassments and losses returning with their original emotional charge when a date cues them.
- Grief: every anniversary would come with full detail. The 2006 case report on Price describes a woman whose remembering dominates her life.
- Grudges: forgiveness is easier when the exact words fade. Total recall keeps the evidence intact.
- Identity: you could not quietly outgrow a past version of yourself, because it would always be one cue away.

Remembering literally everything fails on a counting argument: a lifetime of compressed vision alone, about 1.05 PB, is 2 to 20 times the brain's estimated 50 to 500 TB, and Shannon's source-coding theorem forbids storing it in less. A narrow version exists. About 50 people keep dated autobiographical episodes for decades, but they are not better at most lab memory tasks, still form false memories, and often report the recall as intrusive rather than useful.
What's new
What if next?
- What if you spent a year completely alone? With no one to talk to, your sense of time is the first memory system to drift.
- What if you could upload your mind? If memory lives in 4.7-bit synapses, scanning them all would fill zettabytes.
- What if humans didn't need sleep? You would lose the night shift that decides what you keep.
Questions people ask
Is hyperthymesia a real condition?
Yes. It was first described in a 2006 Neurocase paper and is now studied as highly superior autobiographical memory. It is very rare: the UC Irvine team had identified about 50 cases from several hundred candidates by 2014.
Do people with HSAM have photographic memory?
No. Their strength is dates and personal events. On standard lab tests like digit span or paired words they performed like controls, and Jill Price has said she was poor at rote memorisation.
How much can the human brain store?
Nobody has measured it directly. Using 4.7 bits per synapse (Bartol et al., 2015) and 1014 to 1015 synapses gives roughly 50 to 500 terabytes, an upper bound, since synapses do more than store memories.
Why do we forget things?
Some things are never encoded, some traces weaken, and many are blocked by similar memories at retrieval. Researchers like Richards and Frankland argue forgetting is useful because it keeps decisions flexible and helps you generalise.
Sources
- McGaugh J.L., LePort A.: The Discovery of Super Memories, Scientific American (2014) (source for the Jill Price story and the 50-case figure)
- LePort A.K.R. et al.: Behavioral and neuroanatomical investigation of HSAM, Neurobiology of Learning and Memory (2012) (source for the test table)
- Patihis L. et al.: False memories in highly superior autobiographical memory individuals, PNAS (2013) (source for the false-memory results)
- Parker E.S., Cahill L., McGaugh J.L.: A case of unusual autobiographical remembering, Neurocase (2006) (source for the quote)
- Murre J.M.J., Dros J.: Replication and Analysis of Ebbinghaus' Forgetting Curve, PLoS ONE (2015) (source for the chart data)
- Bartol T.M. et al.: Nanoconnectomic upper bound on the variability of synaptic plasticity, eLife (2015) (source for 4.7 bits per synapse)
- Azevedo F.A. et al.: Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain, J Comp Neurol (2009) (source for 86.1 billion neurons)
- Shapson-Coe A. et al.: A petavoxel fragment of human cerebral cortex, Science (2024) (source for 150 million synapses per mm³)
- Richards B.A., Frankland P.W.: The Persistence and Transience of Memory, Neuron (2017) (source for the decision-making argument)
- Santangelo V. et al.: Altered brain activity during active forgetting in HSAM, iScience (2025)
- Wikipedia: Human brain (accessed 2026) (source for the 1,130 to 1,260 cm³ volume)
- Ministry of Health and Family Welfare: Tele MANAS
More on memory and the mind on the Manthan world page.
