What the Calculator Taught Us About AI in the Classroom

Glasp Research, 2026-11-01. Sources accessed 2026-10-30 unless another date is given. Figures taken from an abstract, a secondary source or the AI reviewer are marked as such. The original measures and comparison rules were committed to version control before the data and the draft; several AI coding rules were added later and are listed under limits.

The fear, the first time

In 1975 the National Advisory Committee on Mathematical Education (NACOME) noted that "for under $20 students can obtain a dependable tool to perform the operations of arithmetic that have long been the focus of elementary school mathematics instruction" [1]. Three years earlier, Texas Instruments' first Datamath had sold for $119.95 [2]. By 1978 calculators cost "approximately one-tenth the price they were four years ago," and "over 80 million calculators have been sold in this country" [3].

Schools were slower. An NCTM survey cited by NACOME found that 83 percent of teachers had never used a hand calculator [1]. A 1977 national survey of 1,177 schools found calculators in 28 percent of K-3 schools and 77 percent of grade 10-12 schools, while only 6 percent of K-3 teachers used them in mathematics classes [3].

The objection was today's: if a machine does the work, students never learn to do it. A survey reported in the Mathematics Teacher found 72 percent of teachers, mathematicians and laypeople opposed giving seventh graders calculators [4, secondary]. In 1986 the textbook author John Saxon warned of a "tremendous potential for destruction" of basic skills [5].

What actually happened, and does that history tell us anything about generative AI?

How the profession answered

Professional bodies backed the tool early, with a reservation about basic facts.

NACOME, 1975: "a calculator should be available for each mathematics student during each mathematics class" from the end of grade 8 at the latest, including on tests [1].
NCTM, An Agenda for Action, 1980: "mathematics programs take full advantage of the power of calculators and computers at all grade levels" (as reprinted in the Christian Science Monitor, 1981) [6].
Cockcroft Report, UK, 1982: secondary pupils should be "taught and allowed to use a calculator," and exam boards should assume one for every candidate "by 1985"; children still needed "a secure grasp of the 'number facts' up to 10 + 10 and 10 x 10" [8].
NCTM Standards, 1989: "appropriate calculators should be available to all students at all times," while "the availability of calculators does not eliminate the need for students to learn algorithms" (quoted by Klein [7], secondary).

The backlash came in the 1990s, folded into the wider "math wars." In California's 1997 standards fight the Superintendent said "we're not even going to let [students] use a calculator before the sixth grade" [7]. In 2000 a Brookings report found that fourth graders who used calculators daily scored lower on NAEP; NCTM's president replied with teacher-reported 1996 NAEP data showing the opposite pattern by class [9]. In 2005 NCTM stressed "balance": "The teacher should help students learn when to use a calculator and when not to" [10].

How assessment changed

The 1990 NAEP mathematics assessment gave students calculators for two of seven sections and classified items as calculator-active, -neutral or -inactive [11]. Today about two-thirds of main NAEP is taken without a calculator [12]. The long-term trend (LTT) version kept "a few" calculator questions until 2004, then dropped them, and does not permit calculators [13][14]. The SAT first allowed calculators in 1994; nearly 95 percent of test takers brought one by November 1997 [15]. The 2016 SAT redesign added a no-calculator section to assess "fluency and number sense" [16], and the digital SAT now embeds a calculator usable throughout the math section [17].

The answer was neither a plain ban nor a plain allowance: some assessments kept a component without the tool, where the basic skill is measured, next to one where the tool is assumed. The arrangement differed by test and was not a fixed end state.

What happened to measured skills

The best public series is the NAEP long-term trend in mathematics, given to national samples of 9-, 13- and 17-year-olds since the early 1970s on a linked scale with a stable core of items (format revised in 2004). Most items were answered without a calculator [13][14], so the series mostly measures broad mathematics performance without the tool, not basic arithmetic alone.

NAEP LTT mathematics, average scale score (standard error in parentheses)

1973*: age 9 219 (0.8), age 13 266 (1.1), age 17 304 (1.1)
1978: age 9 219 (0.8), age 13 264 (1.1), age 17 300 (1.0)
1982: age 9 219 (1.1), age 13 269 (1.1), age 17 298 (0.9)
1986: age 9 222 (1.0), age 13 269 (1.2), age 17 302 (0.9)
1990: age 9 230 (0.8), age 13 270 (0.9), age 17 305 (0.9)
1996: age 9 231 (0.8), age 13 274 (0.8), age 17 307 (1.2)
1999: age 9 232 (0.8), age 13 276 (0.8), age 17 308 (1.0)
2004 (revised format): age 9 239 (0.9), age 13 279 (1.0), age 17 305 (0.7)
2012: age 9 244 (1.0), age 13 285 (1.1), age 17 306 (0.8)

Source: NCES, Digest of Education Statistics 2023, Table 222.85 [18]. *1973 scores are extrapolated and NCES advises caution [19].

Scores did not collapse. Between 1978 and 1999, age 9 rose 13 points and age 13 rose 12, about nine to twelve times the combined standard error. Our check is an approximate screen on rounded estimates, not the NCES procedure, but gaps this large are not in doubt. Age 17 was flat: 304 in 1973, 308 in 1999, 306 in 2012.

We pre-registered a sharper test. If basic skills "faded," the share above the lower NAEP thresholds should fall. If skills "moved up a level," gains should concentrate above the higher thresholds. These are cut points on one scale, not separate tests of computation and reasoning.

Percent of students at or above NAEP LTT performance levels

Age 9, level 200 (beginning skills): 70.4 in 1978, 82.5 in 1999 (+12.1 points), 88.7 in 2012
Age 9, level 250 (numerical operations, beginning problem solving): 19.6 in 1978, 30.9 in 1999 (+11.3 points), 46.7 in 2012
Age 13, level 250: 64.9 in 1978, 78.8 in 1999 (+13.9 points), 84.7 in 2012
Age 13, level 300 (moderately complex procedures and reasoning): 18.0 in 1978, 23.2 in 1999 (+5.2 points), 34.0 in 2012
Age 17, level 300: 51.5 in 1978, 60.7 in 1999 (+9.2 points), 59.7 in 2012
Age 17, level 350 (multistep problem solving and algebra): 7.3 in 1978, 8.4 in 1999 (+1.1 points, not significant), 7.0 in 2012

Source: NCES, Digest of Education Statistics 2013, Table 222.90 [20]; 2012 is the revised format.

Neither signature appears cleanly: the shares above the basic thresholds rose rather than fell, but the gains were spread across thresholds, and the share above 350 for 17-year-olds did not move in 34 years. The honest reading is "no collapse, broad gains in the younger ages." The pre-registered claim that performance "moved toward higher-order problem solving" is not supported.

These series are observational. Calculators spread at the same time as the back-to-basics push, the 1989 Standards and demographic change, so NAEP cannot say the calculator caused the gains. It can say that the predicted collapse is absent. The decline after 2012 (age 13 fell from 285 to 270 by 2025 [18][21]) came long after calculators were established, and we do not attribute it here.

What the controlled studies found

Reviews and meta-analyses summarised studies that compared classes with and without calculators. We read their abstracts, not the full articles, and report no effect sizes.

Suydam, 1982, reviewing 75 studies: 35 percent favoured calculator groups, 44 percent found no difference, 19 percent were mixed, 3 percent favoured non-calculator groups [22].
Hembree and Dessart, 1986 (79 studies): "Except in grade four, use of calculators in concert with traditional mathematics instruction improved average students' basic skills with paper and pencil, both in working exercises and in problem solving" [23]. Hembree's dissertation abstract is blunter: "Sustained calculator usage in Grade 4 appears to damage students' basic skills" [24].
Ellington, 2003 (54 studies) [25]. Her dissertation abstract makes the key split: "When calculators were used during testing, operational and problem solving skills ... realized significant improvement." "When calculators were not used during testing ... the operational skills of average students in grades K-8 and high ability students in all grades were neither helped nor hindered" [26].

These reviews report no average harm to the unassisted skill when the tool was used inside instruction designed around it. And the exception one meta-analysis flagged was basic skills in grade 4, while those skills are still being learned. Our reading, a hypothesis rather than a finding, is that harm is likeliest where the tool replaces practice while the skill is being formed.

The second time: generative AI

Adoption appears far faster. We have no like-for-like series, so this is an illustration across different populations, not a measured duration. Calculators took about two decades to go from mostly unused by teachers (1975) to near-universal on the SAT (1997) [1][15]. Student use of AI for schoolwork reached a majority within about three years of ChatGPT's release in several surveys:

US teens, ChatGPT for schoolwork: 13% to 26% (2023 to 2024), Pew [27]
US teens, any chatbot for schoolwork: 54% (fall 2025), Pew [28]
US high school students, generative AI for schoolwork: 79% to 84% (Jan to May 2025), College Board [29]
UK undergraduates, AI for assessments: 53%, 88%, 94% (2024, 2025, 2026), HEPI [30]

The surveys use different definitions, so they are not one series.

Policy replayed the calculator cycle, compressed. New York City blocked ChatGPT in January 2023 and reversed in May [31]. The International Baccalaureate said in 2023 it would not ban AI tools, comparing them to spell-checkers and calculators [32]. China's May 2025 guidance bars primary pupils from using open-ended generators on their own [33].

Assessment is splitting into lanes again. The University of Sydney's "two-lane" model has a secure, in-person lane and an open lane where AI is scaffolded [34]. US college bookstores report blue-book sales up 30 to 80 percent (the Wall Street Journal, via secondary coverage) [35]. This resembles the NAEP 1990 split, though NAEP split item blocks inside one test while these are institution-wide policies.

The evidence on learning

The calculator literature suggests one question for every study: was the outcome measured with the tool, or without it? We coded the studies we found that way. This is a reading guide, not a systematic review: the search was not reproducible, the studies differ in comparator, outcome and design, and several are preprints or working papers. The list below counts 11 study arms from 10 publications. "Null" means no reported difference or an interval including zero; "not established" means no outcome was shown to be measured without the tool. For most study arms we read only the abstract, so a direction is usually the sign of the estimate, not a tested result.

Unrestricted chatbot: 2 negative, 2 null, 0 positive, 0 not established on the unassisted outcome
Guarded tutor: 0 negative, 2 null, 2 positive, 1 not established on the unassisted outcome
Support for the human teacher or tutor: 0 negative, 0 null, 1 positive, 0 not established on the unassisted outcome
Course-integrated generative AI tutor, design unclear: 0 negative, 0 null, 0 positive, 1 not established on the unassisted outcome

The clearest studies:

Bastani et al., PNAS 2025, a field experiment with nearly a thousand Turkish high school math students. With a plain GPT-4 interface, practice scores rose 48 percent, but the later exam without AI fell 17 percent relative to students who never had access (statistically significant). A tutor version with guardrails raised practice scores 127 percent and "largely mitigated" the exam harm; its exam estimate was statistically indistinguishable from the control group's [36].
Stromberg, Lei and Wu, 2026 (working paper), 26,811 Chinese students in grades 7 to 12, difference in differences, not randomised: homework scores +18 percent, monthly exam scores -20 percent within six months [38].
Liu et al., 2026 (working paper), RCT with 2,379 US undergraduates: course-integrated AI tutoring lowered final grades by 0.37 SD [37]. Whether the grade was earned without the tool is not established, so it is not counted as negative.
De Simone et al., World Bank 2025, RCT in Nigeria: a teacher-guided GPT-4 after-school program raised English by about 0.23 SD (abstract) [39]. The AI reviewer reports heavy attrition from the full paper (1,328 randomised, 759 at the final assessment).
Henkel et al., 2024, Rori math tutor in Ghana: effect size 0.37 on math growth (preprint v1 abstract) [40].
Wang et al., Tutor CoPilot (preprint): AI that coaches human tutors raised student topic mastery by 4 percentage points, and by 9 for lower-rated tutors [41].
LearnLM and Eedi, 2025 (exploratory RCT, 165 UK students): 66.2 percent on novel problems with the AI tutor vs 60.7 percent with a human tutor, with a 95 percent credible interval of -1.4 to +12.4 points for the difference [42]. We code it null.

Both positive guarded-tutor studies compare an added programme with business as usual, so extra instruction time is part of the effect. Two lab studies found no overall difference in learning without the tool: Fan et al., who coined "metacognitive laziness," found better essays with ChatGPT but no significant difference in knowledge gain or transfer [44], and Lehmann et al. found no overall effect [43]. A widely cited 2025 meta-analysis was retracted on 22 April 2026 [45]; we do not use it. A March 2026 meta-analysis of 35 experiments reports a pooled effect on learning of g = 0.670 [46] (95 percent confidence interval 0.495 to 0.844, with very high heterogeneity, both per the AI reviewer), but its abstract does not separate outcomes measured with the tool from those without it.

Where the analogy holds and where it breaks

It holds on design. The calculator reviews attributed benefits to use "in concert with traditional mathematics instruction" [23], not to availability alone. The clearest AI study points the same way: in Bastani et al.'s experiment the plain chatbot lowered the later exam taken without AI, while the guarded tutor "largely mitigated" that harm [36]. The other studies do not contradict this but are too varied to establish it, so the pattern is provisional.

It partly holds on assessment. In both eras some systems created a component without the tool next to one with it. Not every system did, and the calculator split later partly re-merged.

It breaks on what the tool does. A basic calculator automated computation and left the problem solving to the student, which is why "moving up a level" was plausible. Generative AI can also produce the reasoning, explanation and writing in many text-based tasks. The calculator harm flagged in the reviews was where the tool did the very skill being learned [24]. Because generative AI can do a much larger share of what students are learning in text, we expect, as a hypothesis, that this exception becomes common under unrestricted use. Bastani et al. (randomised) and Stromberg et al. (observational) both found higher scores with the tool and lower without it.

It may break on speed. Use reached a majority within about three years, while calculator policy and testing adjusted over about two decades. The AI research base is thin: a few large trials, several of them working papers, and the one pooled estimate we found is not split by whether the tool was on the test.

What the calculator teaches

What fades may be what nobody practises. Arithmetic did not collapse while curricula kept number facts and some tests kept a no-calculator lane. We cannot show that these prevented fading; it is a plausible reading.
"Moving up a level" is a design goal, not an automatic effect. NAEP shows broad gains, not a jump at the top. With AI the level above is also automatable, so a curriculum has to choose what students still do themselves.
Expect a secure lane for a long time. The SAT went from no calculator, to calculator allowed (1994), to a split (2016), to a calculator built in. As a forecast, not a finding, we expect a similar arc for AI.
Ask of every study whether the tool was on the test. It sorted the calculator literature, and it is the most useful single question to ask of the AI studies today.

Departures from the plan, and limits

Of the pre-registered calculator hypothesis, "no collapse" survives and "moved toward higher-order problem solving" is not supported. Whether any meta-analysis found significant harm to computation without the calculator is unresolved: we did not see effect sizes. Planned effect sizes and classroom-use rates by year were not collected.
The AI coding rules (exclusions, "not established", interval-including-zero as null) were added after the data were seen, during review. Excluded: a neural and self-report study (Kosmyna et al.), a feedback tool that is not generative AI (Darvishi et al.) and the retracted meta-analysis.
The AI set is a selection, not weighted or rated for risk of bias; intervals were read only for Bastani et al. and LearnLM/Eedi. Liu, Stromberg and De Simone are working papers; the Wang, Lehmann, Kosmyna, Henkel and LearnLM/Eedi papers were read as preprints.
The full texts of Hembree and Dessart and of Ellington needed a login. The 1980 and 1989 NCTM wording is known through a newspaper reprint and Klein. Whether the pre-2004 LTT calculator items entered the reported scores is not known.

References

[1] CBMS/NACOME (1975). Overview and Analysis of School Mathematics, Grades K-12. ERIC ED115512. https://files.eric.ed.gov/fulltext/ED115512.pdf

[2] Computer History Museum, TI Datamath 2500 (1972). https://www.computerhistory.org/revolution/digital-logic/12/279/1459

[3] Suydam, M. N. (1979). Calculator review, Calculator Information Center. ERIC ED171573. https://files.eric.ed.gov/fulltext/ED171573.pdf

[4] Banks (2011). Master's thesis on the history of calculator attitudes in US schools. ERIC ED525547. https://files.eric.ed.gov/fulltext/ED525547.pdf

[5] Rowe, J. (1986, May 9). The great calculator debate. Christian Science Monitor. https://www.csmonitor.com/1986/0509/dcalc-f.html

[6] Christian Science Monitor (1981, March 23), reprint of NCTM An Agenda for Action recommendations. https://www.csmonitor.com/1981/0323/032305.html

[7] Klein, D. (2003). A brief history of American K-12 mathematics education in the 20th century. https://www.csun.edu/~vcmth00m/AHistory.html

[8] Cockcroft Committee (1982). Mathematics Counts. HMSO. https://www.educationengland.org.uk/documents/cockcroft/cockcroft1982.html

[9] Stiff, L. V. (2001). Making calculator use add up. NCTM President's Message. https://www.nctm.org/News-and-Calendar/Messages-from-the-President/Archive/Lee-V_-Stiff/Making-Calculator-Use-Add-Up/

[10] Cavanagh, S. (2005, June 7). NCTM elaborates on position on the use of calculators in classrooms. Education Week. https://www.edweek.org/teaching-learning/nctm-elaborates-on-position-on-the-use-of-calculators-in-classrooms/2005/06

[11] NCES (1991). NAEP 1990 Trial State Assessment, mathematics report for California. ERIC ED330550. https://files.eric.ed.gov/fulltext/ED330550.pdf

[12] NCES. What does the NAEP mathematics assessment measure? https://nces.ed.gov/nationsreportcard/mathematics/whatmeasure.aspx

[13] NCES (2005). NAEP 2004 Trends in Academic Progress (NCES 2005-464), Appendix A. https://nces.ed.gov/nationsreportcard/pdf/main2005/2005464.pdf

[14] NCES. More about the long-term trend assessment. https://nces.ed.gov/nationsreportcard/ltt/moreabout.aspx

[15] Scheuneman and Camara (2002). College Board Research Notes RN-16. ERIC ED562590. https://files.eric.ed.gov/fulltext/ED562590.pdf

[16] Loewus, L. (2014, March 14). The new SAT: What will it look like for math? Education Week. https://www.edweek.org/teaching-learning/the-new-sat-what-will-it-look-like-for-math/2014/03

[17] College Board. SAT Suite calculator policy. https://satsuite.collegeboard.org/digital/what-to-bring-do/calculator-policy

[18] NCES (2023). Digest of Education Statistics, Table 222.85. https://nces.ed.gov/programs/digest/d23/tables/dt23_222.85.asp

[19] NCES / NAEP. About the 2025 long-term trend assessment. https://www.nationsreportcard.gov/ltt/2025/about/

[20] NCES (2013). Digest of Education Statistics, Table 222.90. https://nces.ed.gov/programs/digest/d13/tables/dt13_222.90.asp

[21] NAGB (2026, June 10). Press release, NAEP long-term trend 2025. https://www.nagb.gov/content/dam/nagb/en/documents/newsroom/press-releases/2026/press-release-ltt-2025-final.pdf

[22] Suydam, M. N. (1982). The Use of Calculators in Pre-College Education: Fifth Annual State-of-the-Art Review. ERIC ED220273. https://files.eric.ed.gov/fulltext/ED220273.pdf

[23] Hembree, R., and Dessart, D. J. (1986). Effects of hand-held calculators in precollege mathematics education: A meta-analysis. Journal for Research in Mathematics Education, 17(2), 83-99. https://doi.org/10.2307/749255 (abstract via ERIC EJ336469)

[24] Hembree, R. (1984). Doctoral dissertation, University of Tennessee. https://trace.tennessee.edu/utk_graddiss/12884/

[25] Ellington, A. J. (2003). A meta-analysis of the effects of calculators on students' achievement and attitude levels in precollege mathematics classes. Journal for Research in Mathematics Education, 34(5), 433-463. https://doi.org/10.2307/30034795

[26] Ellington, A. J. (2000). Doctoral dissertation, University of Tennessee. https://trace.tennessee.edu/utk_graddiss/8272/

[27] Pew Research Center (2025, January 15). About a quarter of U.S. teens have used ChatGPT for schoolwork. https://www.pewresearch.org/short-reads/2025/01/15/about-a-quarter-of-us-teens-have-used-chatgpt-for-schoolwork-double-the-share-in-2023/

[28] Pew Research Center (2026, February 24). How teens use and view AI. https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/

[29] College Board (2025, October 6). New research: majority of high school students use generative AI for schoolwork. https://newsroom.collegeboard.org/new-research-majority-high-school-students-use-generative-ai-schoolwork

[30] HEPI (2025, 2026). Student Generative AI Survey 2025; Student Generative AI Survey 2026. https://www.hepi.ac.uk/2025/02/26/student-generative-ai-survey-2025/ and https://www.hepi.ac.uk/reports/student-generative-ai-survey-2026/

[31] NBC News (2023, May 18). ChatGPT ban dropped by New York City public schools. https://www.nbcnews.com/tech/chatgpt-ban-dropped-new-york-city-public-schools-rcna85089

[32] International Baccalaureate (2023). Statement from the IB about ChatGPT and artificial intelligence in assessment and education. Wording confirmed by the AI reviewer on 2026-11-01; our own fetch was refused (403). https://www.ibo.org/news/news-about-the-ib/statement-from-the-ib-about-chatgpt-and-artificial-intelligence-in-assessment-and-education/

[33] China Daily (2025, May 16). Guideline to regulate use of artificial intelligence in schools. https://govt.chinadaily.com.cn/s/202505/16/WS6826de33498eec7e1f736de6/guideline-to-regulate-use-of-artificial-intelligence-in-schools.html

[34] University of Sydney, Teaching@Sydney, two-lane assessment FAQ (2024, updated 2026). https://educational-innovation.sydney.edu.au/teaching@sydney/?p=21264

[35] Entrepreneur (2025). College professors turn back to blue books to combat ChatGPT (relaying the Wall Street Journal). https://www.entrepreneur.com/business-news/college-professors-turn-back-to-blue-books-to-combat-chatgpt/492450

[36] Bastani, H., et al. (2025). Generative AI without guardrails can harm learning. PNAS, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122

[37] Liu et al. (2026). Effects of course-integrated AI tutoring. EdWorkingPaper 26-1598. https://edworkingpapers.com/ai26-1598

[38] Stromberg, Lei, and Wu (2026). The Generative AI Learning Penalty. CEPR Discussion Paper 21577. https://econpapers.repec.org/RePEc:cpr:ceprdp:21577

[39] De Simone et al. (2025). From chalkboards to chatbots. World Bank Policy Research Working Paper 11125. https://econpapers.repec.org/RePEc:wbk:wbrwps:11125

[40] Henkel et al. (2024). Effective and scalable math support (Rori). arXiv:2402.09809, version 1 (figures cited from the v1 abstract). https://arxiv.org/abs/2402.09809v1

[41] Wang et al. (2024, revised 2026). Tutor CoPilot. arXiv:2410.03017. https://arxiv.org/abs/2410.03017

[42] LearnLM Team, Google, and Eedi (2025). AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms. arXiv:2512.23633 (interval read in the v1 HTML text, 2026-11-01). https://arxiv.org/abs/2512.23633

[43] Lehmann, Cornelius, and Sting (2024, revised 2025). AI meets the classroom. arXiv:2409.09047. https://arxiv.org/abs/2409.09047

[44] Fan et al. (2025). Beware of metacognitive laziness. British Journal of Educational Technology, 56(2). https://doi.org/10.1111/bjet.13544

[45] Retraction Note: The effect of ChatGPT on students' learning performance, learning perception, and higher-order thinking: insights from a meta-analysis (Wang, J., and Fan, W.). Humanities and Social Sciences Communications, 13 (2026, April 22). Notice: https://doi.org/10.1057/s41599-026-07310-z (retracts https://doi.org/10.1057/s41599-025-04787-y)

[46] Wu, X., Zhu, P., Zhang, J., Yin, M., and Wang, Y. (2026, March 26). ChatGPT's impact on student learning outcomes: a meta-analysis of 35 experimental studies. Humanities and Social Sciences Communications, 13, 684. Metadata via Crossref and abstract via Semantic Scholar, 2026-11-01; full text not read. https://doi.org/10.1057/s41599-026-07019-z

AI-use disclosure

This post was researched and written by an AI model (Anthropic's Claude) in an automated Glasp research run. AI sub-agents searched and read public sources and recorded the source URL used for each number; where a figure came from a search snippet, an abstract or a secondary source, the records say so. The draft was then reviewed in several rounds by a second AI system (OpenAI Codex) acting as referee, and revised by Claude in response; "the AI reviewer" in the text refers to that system, and figures attributed to it were not confirmed by us. The selection of studies, their coding and the quotes have not been checked by a human. Prepared for publication by Glasp.