Super Intelligence master guide › Energy, work, learning and safety series › Article 0014
AI-assisted research can shorten the work of organising questions, locating material, comparing options and preparing a first report. The useful result is a clearer path from a question to an evidence-based judgment. Speed matters when the output preserves sources, assumptions and uncertainty, and when the reader still understands what the report means.
For adults, students and parents, the most effective starting point is a research brief. Describe what you want to understand, which options or ideas you want compared and what would count as a useful answer. A single vague request for “everything about a topic” can produce abundant text without resolving the question that prompted the research.
This guide explains comparative research, source checking, rates of change and chart interpretation through hypothetical examples. It does not promise that a complex investigation can always be completed in minutes. Preparation time, verification and learning are different parts of the process, and all belong in an honest account of the benefit.
The series uses “Super Intelligence” as its title, while artificial superintelligence remains hypothetical. Current AI systems can assist with particular research activities and still produce errors. A fast, polished report is not evidence that the system has achieved general superior judgment or that its conclusions should be accepted without examination.
Research begins with a question, not a report format
A question gives research direction. It tells the researcher what information matters, which comparisons are relevant and what uncertainty needs to be reduced. A report format organises the result, but should follow the question rather than substitute for it.
“Compare three study approaches for a student with limited evening time” is different from “write about study methods.” The first request defines a decision context. It suggests questions about duration, requirements and the kind of practice involved. The second may produce a broad overview without helping the student choose a workable approach.
The same distinction applies at work. A team comparing tools needs criteria connected to its task. A list of attractive features may not explain whether any tool fits the actual workflow. The research brief should state the activity the comparison is intended to support.
Ask the whole set of relevant questions
A comparative brief can include several connected questions. What does each option require? What information is missing? How do costs or rates change over time? Which differences matter for the intended decision? Asking the set together can help the assistant organise a coherent investigation.
This does not mean every possible question deserves equal attention. Separate the questions essential to the decision from those that provide background. That priority helps keep the report focused and gives the reviewer a reason to check some claims more deeply than others.
Define a useful stopping point
Research can continue indefinitely if the task has no completion condition. A practical brief might ask for a first comparison using specified sources, an explanation of the main uncertainty and a list of questions that remain unresolved. That produces a result the user can review.
A stopping point is not a declaration that the topic is fully understood. It is the boundary of the current investigation. Defining it prevents a first report from being mistaken for a comprehensive final answer and makes the next step more deliberate.
Build a comparison that treats options fairly
A useful comparison applies shared criteria to each option. If one option is described by cost and another by convenience, the reader cannot easily judge the difference. Consistent criteria make the structure visible and expose gaps in the evidence.
Suppose a hypothetical comparison examines three courses. One provides total duration, one states weekly hours and one says only that the schedule is flexible. The assistant should not convert these descriptions into an invented common figure. It can preserve what is known and mark what would be needed for a fairer comparison.
Comparable definitions matter
A word such as “completion” can mean attendance, passing an assessment or receiving a certificate. A word such as “cost” might include only a listed fee or also necessary equipment. Research should identify the meaning used by each source before treating the values as equivalent.
This is a practical way to prevent a polished table from hiding differences. A comparison can include a note explaining that two figures use different boundaries. The note may be more important than another decimal place in the table.
Include the user’s constraints
An option that performs well in general may be unsuitable for the user’s time, budget or available equipment. The brief should therefore distinguish general characteristics from fit for the particular task. This helps the report support a decision without claiming a universal winner.
For students, the skill is to explain why a criterion matters. If limited evening time is the constraint, a course’s total length may matter less than the required weekly commitment. Research becomes more useful when the comparison follows the real problem rather than a standard feature checklist.
Sources should support individual claims
A report is easier to review when important claims are connected to the material that supports them. A long list of links at the end may be useful for further reading, but it does not show which source establishes a particular conclusion.
Primary material is often a good starting point for a technical or institutional claim: official documentation, a research paper, an original dataset or the organisation’s own description of a policy. The source still needs to be examined for scope and date. Being primary does not make every interpretation of it correct.
Read what the source actually says
A source may describe a proposal, an observed result or a forecast. These are different kinds of statements. The report should preserve that distinction. A target should not become a completed achievement, and a result from one context should not become a universal claim.
Checking also includes confirming that a link reaches the intended material and that the material supports the claim near it. A plausible title is insufficient. The evidence should be available in the source itself, not inferred from the appearance of a citation.
A generated citation can be wrong
NIST’s generative AI profile identifies confident erroneous content and misleading generated citations among relevant risks. This supports verifying important claims and references rather than relying on an authoritative tone. NIST: Generative AI profile.
The practical response is a reviewable report. Ask the assistant to identify the basis of each important statement, mark uncertainty and distinguish missing evidence from a negative conclusion. A source not found is not proof that a claim is false; it is a limit on what the current investigation establishes.
A demanding application is healthcare claims traced through the evidence chain; literature relevance, validated performance and clinical benefit are different claims requiring appropriate evidence.
Rates of growth require careful arithmetic
Comparing rates of change can reveal a pattern, but the starting value, period and definition must be visible. A percentage alone is easy to misread. It describes a change relative to a base, not necessarily the largest absolute increase.
Suppose a hypothetical measure rises from 100 to 120 over a year. The increase is 20, and the percentage increase is 20 percent. Another measure rises from 1,000 to 1,100. Its absolute increase is 100, while its percentage increase is 10 percent. The first grows faster in percentage terms; the second adds more in absolute terms.
Neither statement is universally more important. The relevant one depends on the question. A small organisation may care about relative expansion, while a supply plan may care about the quantity added. A good report states both when the distinction affects the interpretation.
Align the period
A change over one month should not be compared directly with a change over one year without an explanation. Different periods can create a misleading ranking. The report should show the interval for each figure and avoid assuming that a short-period rate continues unchanged.
For a hypothetical example, a measure rises rapidly in its first month and then stabilises. Multiplying the first month’s change across a year would describe a scenario, not an observed annual result. That distinction should appear explicitly if such a calculation is included.
Percentage points and percentages differ
If a share rises from 40 percent to 44 percent, it rises by four percentage points. Relative to the original share, the increase is ten percent. These descriptions answer different questions and should not be swapped.
Students can practise the distinction with simple examples. The skill is not merely remembering terminology; it is understanding the denominator. A report comparing rates should explain which quantity changed and what the percentage is relative to.
Worked example: comparing two growth patterns
Imagine a hypothetical school club examining participation in two activities. Activity A has 20 participants in the first term and 30 in the next. Activity B has 100 participants and then 120. The club wants to understand which activity is growing and what support may be needed.
Activity A adds ten participants and grows by fifty percent. Activity B adds twenty and grows by twenty percent. A headline saying A is growing faster is accurate in relative terms, but does not describe the larger additional number that B needs to accommodate.
The club next asks whether the counts use the same definition. If A counts unique participants while B counts attendances across several sessions, the figures are not directly comparable. The calculation can be correct while the interpretation remains weak. The report should identify the mismatch before producing a ranking.
Now consider the period. If one count covers a term and the other a month, the comparison needs another qualification. The club might obtain aligned records or explain that the current evidence cannot establish a fair rate comparison. Marking this limitation is part of sound research.
The final report could show absolute change, relative change and the definition of each count. It could then explain that support planning needs the number of additional participants, while outreach analysis might use relative growth. The same data supports different decisions when the question changes.
This example shows what an AI-assisted report should do well: organise the calculations, preserve definitions and explain the result in relation to the user’s purpose. A colourful chart without those steps could make the comparison more convincing while leaving it less reliable.
Charts should clarify the reasoning
A chart can make a pattern easier to see, but its design affects interpretation. The type of chart, labels, scale and included data should follow the research question. Decorative variety is not a reason to choose a chart that obscures the comparison.
A line chart can show a sequence over time. A bar chart can compare categories. A table may be preferable when exact values and qualifications are central. The appropriate form depends on what the reader needs to examine, and a report can combine a simple chart with a small supporting table.
Make units and boundaries visible
A chart should state what is measured, over which period and for which population or set of cases. Without that context, the reader may supply an assumption the data does not support. A title saying “growth” is less useful than one naming the measure and interval.
The Office for National Statistics’ chart guidance explains why a bar chart’s numerical axis should start at zero when bar length represents magnitude. This is one example of how a visual choice can change the apparent size of a difference. ONS: Axes and gridlines.
Do not let the chart outrun the evidence
A smooth line can suggest continuous information even when only a few observations exist. A precise-looking label can imply more certainty than the measurement supports. The report should explain what was actually observed and what has been interpolated or calculated.
For a hypothetical case with only three recorded periods, the chart can show those periods without pretending to know the values between them. If a scenario is plotted alongside observations, identify it clearly. Presentation should preserve the status of the information rather than blur it.
Choose a chart that supports the decision
If the question concerns absolute additional capacity, a chart of percentage growth alone may distract from the relevant quantity. If the question concerns composition, a display that compares totals may leave the share unclear. Start with the question and choose the visual form afterward.
This is a useful habit for learners. Explain in one sentence what the chart should help the reader see. Then check whether the chosen form actually makes that point easier to assess. If it does not, a simpler chart or table may be better.
Worked example: turning a list of questions into a report
Consider a hypothetical adult comparing three ways to learn a new skill while working full time. The research brief asks about weekly commitment, required background, cost boundaries and the kind of practice included. It also asks which details remain unclear.
The assistant begins by organising the questions into a shared comparison structure. It locates suitable material within the task’s permitted scope and records the source for each important entry. Where a description omits a required detail, it marks the gap.
Suppose one option describes flexible access but provides no estimate of weekly effort. The report should not conclude that it is automatically easiest to fit around work. Flexibility concerns scheduling; effort concerns the amount of work. The distinction directly affects the user’s decision.
Next the report separates facts from fit. The stated duration and requirements belong in the evidence section. The conclusion about suitability depends on the user’s time and starting knowledge. The report can explain that conditional relationship instead of declaring one option best for everyone.
The reviewer checks the highest-impact details and confirms that the sources support them. They also inspect whether any important criterion was omitted. This review may take longer than reading the report, but it makes the research useful for the decision.
The final result includes a comparison, a short explanation of the trade-offs and the questions to resolve before committing. The report is complete for its defined stage even if the user still needs one missing detail. Research has reduced uncertainty and prepared a concrete next step.
Faster preparation and deeper learning are different outcomes
An assistant can reduce the effort of assembling a first account of a topic. Learning requires the reader to understand the relationships, recognise uncertainty and apply the ideas. A report that arrives quickly creates an opportunity for learning, but does not guarantee it.
UNESCO’s guidance for generative AI in education and research emphasises human agency and meaningful use. For a learner, that supports asking what understanding remains with the student after assistance rather than treating output completion as the educational goal. UNESCO: Generative AI in education and research.
Read actively rather than accept passively
An active reader can ask why a conclusion follows, which assumption matters and what evidence would change it. These questions turn a report into a starting point for thinking. They are especially useful when the report’s presentation is polished enough to make uncertainty less visible.
A practical exercise is to explain the main conclusion without looking at the report. Then identify one limitation and one question that remains. If the reader cannot do this, more time spent with the reasoning may be useful before relying on the conclusion.
Use saved time deliberately
If preparation becomes faster, decide how to use the time saved. It might support checking a central source, exploring an alternative explanation or attempting an application. Without that choice, faster output can simply create more reading material.
For students, a generated explanation can be followed by a related problem solved independently. For adults, a comparison can be followed by examination of the option that appears most promising. The next activity should match the reason for the research.
The educational consequence depends on saved research time used for independent reasoning; receiving a report should create an opportunity to explain, challenge and apply the reasoning rather than merely accept it.
Worked example: a student investigates an energy question
Imagine a student asking whether batteries can supply a computing facility through a four-hour period. The research brief names the hypothetical load and asks for an explanation of power, energy and missing assumptions. It does not ask the assistant to identify a real product to buy.
Suppose the load is five megawatts. In an idealised calculation, four hours requires twenty megawatt-hours. The report should distinguish this energy quantity from the five-megawatt discharge rate. It should also explain that a real design would require additional operating details rather than treat the arithmetic as a complete specification.
The student then checks the explanation using the source definition and reproduces the calculation. They ask what changes if the load doubles or the required period becomes shorter. This turns the report into a learning exercise and tests whether the distinction is understood.
Next the student identifies what the simplified calculation leaves out. They may ask about usable capacity, charging or the role of the battery in the wider system. These are questions for further research, not facts the original arithmetic established.
The task is successful when the student can explain the relationship and the limits of the model. A longer generated answer would not necessarily improve that outcome. The useful research process combines a clear question, a reviewable explanation and active application.
The connected infrastructure topic is developed in AI demand and the future of energy. Following that link makes sense when the learner wants to extend the question from units to investment and supply.
Verification should follow the claim’s importance
A report may contain background details and central claims. The central claims determine the conclusion and deserve priority in review. If a recommendation rests on a cost, requirement or deadline, verify that item carefully before spending time polishing minor wording.
This does not mean ignoring the rest. It means allocating attention in proportion to the role a statement plays. A useful review identifies which error would change the decision and which would merely affect presentation.
Reproduce the calculation
When a conclusion depends on arithmetic, inspect the inputs, formula and units. A result can be numerically correct while using the wrong period or definition. Reproducing the steps helps reveal both calculation errors and interpretation errors.
For a simple percentage change, write the original value, new value, difference and denominator. For a total, identify which items are included. The report should make these steps accessible enough that another reader can check them without guessing.
Look for another explanation
A comparison can show that two measures change together without establishing why. An assistant may propose a causal story that goes beyond the evidence. The reviewer should ask what else could explain the pattern and whether the source actually supports the mechanism.
In a hypothetical study of participation, growth might reflect a changed counting method rather than increased interest. In a product comparison, a lower price might omit a required item. Alternative explanations are not objections for their own sake; they test whether the conclusion follows from the information available.
Reliable inquiry depends on truth that remains independent of a preferred conclusion; the desired outcome can guide a question without deciding what the evidence is allowed to show.
Preserve unresolved questions
An unresolved question should appear near the conclusion it affects. Hiding limitations at the end can make the main recommendation seem stronger than it is. The reader should know whether a missing fact could change the result.
A useful report can say that two options are comparable on the supplied information while a third needs clarification. That is more informative than forcing a complete ranking. Research reduces uncertainty honestly, including by showing where it remains.
The practical mechanism is verification matched to the kind of claim; calculations, changing facts and judgments need different checks rather than a single confidence impression.
Worked example: two sources disagree
Imagine a hypothetical report comparing three services. One official page lists a requirement, while another document from the same organisation describes a different requirement. The assistant finds both. A complete-looking answer could quietly choose one, but that would hide the uncertainty from the reader.
The first step is to identify the dates and scope. One document might concern a different programme or an earlier version. If the source itself resolves the difference, the report can explain that resolution. If it does not, the report should preserve the conflict and state what information would settle it.
The next step is to assess whether the disagreement affects the decision. A minor wording difference may have little consequence. A different eligibility requirement could determine whether the option is suitable at all. The report should place the limitation where the reader makes the relevant judgment.
Suppose the user needs only an initial shortlist. The report might retain the option with a note that its requirement needs confirmation. If the user is ready to commit, that unresolved item becomes more important. The same evidence can support different stopping points depending on the task.
A useful research record would list the claim, the two sources, the possible explanation and the status. This makes later review efficient. Another reader can see the actual problem rather than reconstruct why the assistant selected one statement.
The example shows why uncertainty should not be treated as a defect to conceal. A report that accurately identifies a conflict can move the work forward by making the next question precise. That is a substantive contribution to research even when it prevents an immediate final ranking.
Keep a small evidence ledger
For a longer investigation, a concise record of important claims can support continuity. Each entry can identify the claim, supporting material, calculation where relevant and unresolved limitation. The record should focus on statements that affect the conclusion.
This is especially useful when several questions connect. A changed definition in one section may affect a calculation elsewhere. Keeping the dependency visible makes revision easier and reduces the chance of retaining a conclusion built on an outdated premise.
The ledger is an aid to thinking rather than another report to produce for its own sake. If it helps the reviewer find the evidence and understand what remains open, it is doing useful work. If it repeats every sentence without improving review, a simpler record may be sufficient.
Measure the time benefit fairly
The assisted process includes writing the brief, obtaining material, generating the report, checking it and understanding it. If only generation time is counted, the time-saving claim will be incomplete. The right comparison uses the same intended outcome before and after assistance.
Suppose a hypothetical task previously takes three hours. The assisted process takes twenty minutes to prepare, ten to generate, forty to verify and thirty to understand and refine. The total is one hundred minutes, not ten. Under those assumptions, the process still saves eighty minutes.
The example shows why honest measurement can support optimism more effectively than an exaggerated claim. It identifies the actual source of the benefit and makes the remaining work visible. It also reveals where a better brief or clearer source structure might help further.
Compare equivalent quality
A rough initial overview should not be compared with a carefully verified final report as though they are the same deliverable. The speed difference may partly reflect a difference in standard. State what each process produced and what review remains.
This is particularly useful for students. A completed page may satisfy a formatting requirement without demonstrating understanding. Compare the activity that matters: a reliable explanation, a usable decision or a learned skill. The standard should follow the purpose.
Common mistakes in AI-assisted research
A vague request can produce a broad answer that never resolves the real question. Define the intended decision or understanding, then specify the comparisons needed. This gives the report a direction and a useful stopping point.
Another mistake is treating sources as decoration. A citation should support the nearby claim, and the reviewer should inspect central sources. An impressive list of references does not establish that the report interpreted them correctly.
A third mistake is mixing observed results with conditional scenarios. Both can be useful, but their status must be clear. A projection based on an assumed rate is different from a measured future outcome.
A fourth mistake is comparing rates with different periods or denominators. Align the quantities or explain why they cannot yet be compared. Precision in arithmetic cannot repair incompatible definitions.
Finally, a reader can mistake a clear report for personal understanding. Explain the conclusion, test an application and identify a limit. These activities help convert fast preparation into durable knowledge.
A practical research workflow
Begin with a question and the purpose of the result. Identify the options or concepts being compared and the essential criteria. Decide whether the task is an initial overview, a source-based comparison or a preparation stage for a consequential decision.
Ask for a structure that separates evidence, calculations, interpretation and unresolved questions. This makes the report easier to review and prevents unsupported conclusions from blending into sourced material. The structure should serve the topic rather than become an inflexible template.
Examine important sources and reproduce central calculations. Check definitions, periods and units. Revise the conclusion if the evidence does not support the initial interpretation. The ability to update the answer is part of good research.
Then use the result. Make the appropriate next decision, attempt a learning exercise or identify the next question. A report becomes valuable through that connection to action or understanding. Generating more reports without using them can recreate an information burden at a higher speed.
For the delegation side, read work you can delegate. For the distinction between quick answers and learning, continue with faster answers and deeper learning.
Frequently asked questions about AI-assisted research
Can AI complete research instantly?
It can sometimes prepare an initial output quickly, depending on the task and tool. Research also includes finding suitable evidence, checking claims and understanding the result. Those activities should not disappear from the account of completion.
Define the deliverable and count the whole process. A fast first report can be useful without being a fully verified final investigation.
What should I include in a research prompt?
State the question, purpose, relevant options, comparison criteria and permitted material. Ask for missing information and uncertainty to be made visible. Define the form of the result only after clarifying what it should help you understand.
You do not need an elaborate formula. A plain brief with a clear task is often more useful than a long instruction that leaves the goal vague.
How do I check whether a citation is trustworthy?
Open the material, confirm that it is the intended source and inspect whether it supports the nearby claim. Check its date, scope and the status of the statement. A source can describe a forecast rather than an observation.
For central claims, keep enough traceability to review the interpretation. The citation’s appearance is not the evidence; the supporting content is.
Should every report include charts?
Only when a chart helps answer the question or makes the evidence easier to examine. A table or connected explanation may be better when exact values, definitions or qualifications matter most.
Choose the visual form after identifying the message and data. Charts should clarify reasoning rather than make an incomplete comparison seem complete.
How can parents help children use AI research well?
Ask the child to explain the conclusion, identify the source of a central claim and apply the idea to a new example. Help them distinguish what was observed from what was assumed. These habits support understanding and responsible use.
Return to the Super Intelligence guide for the broader connections among research, software, delegation, energy and the practical responsibilities of using AI.
Previous: 0013 — The Economics of Software Abundance · Next: 0015 — AI Literacy for Work and Everyday Life