Evidence checked October 31, 2026. Human studies are described at their actual scale and dates. Numerical scenarios are hypothetical. This article does not recommend implantation or participation in a clinical trial.
1. The more consequential milestone happens at home
A June 15, 2026 UC Davis report described a man with ALS using an investigational brain-computer interface at home for communication and computer access without researcher support. The report distinguished accuracy during controlled testing from the participant's ratings of sentences during everyday use. The research institution's account concerned one participant, not a broadly available clinical service or proof that the same result will occur for everyone.
That distinction makes the achievement more informative, not less. Moving a cursor during a demonstration establishes a capability under the conditions shown. Sustained use outside a research session asks whether the capability can become a dependable part of someone's life. The second question includes setup, errors, interruptions, changing signals, and the ability to communicate something the experimenter did not choose in advance.
A brain-computer interface, or BCI, is not just an implant. It is a measurement and control system connecting recorded neural activity to an output. Its usefulness depends on the sensor, decoding software, interface, external equipment, and support around the user. A failure anywhere in that chain can interrupt the function the person needs.
The next stage of evaluation therefore requires more than a visually impressive moment. It requires a description of what the user can accomplish, with what assistance, how reliably, for how long, and at what cost in risk and effort. Those questions provide a more serious standard for progress than either unqualified excitement about mind reading or reflexive dismissal of small early studies.
2. A successful task needs a denominator
A video can show a cursor reaching a target or a sentence appearing on a screen. It rarely shows how many attempts preceded the successful one, how long setup took, or how often the system was unavailable. Those omitted quantities are not peripheral to performance. They define the conditions under which the displayed success can be expected.
A hypothetical cursor study might report ninety successful selections in one hundred attempted trials. That is useful information if the targets, timing, and failure criteria are clear. It means something different if the hundred trials were selected from a larger set after difficult sessions were excluded. It also means something different if each selection required a technician to reset the system.
The denominator should match the claim. For a technical feasibility claim, completed laboratory trials may be appropriate. For independent daily use, the relevant denominator may include days the participant wanted to use the device but could not, sessions that failed to start, and tasks abandoned because correction became burdensome. A headline rate calculated only from successful operation can miss the practical reliability problem.
Reporting this information does not require turning every paper into an exhaustive diary. It requires separating peak performance from typical performance and availability. A system may be excellent when active yet difficult to make available. Another may have a lower peak rate but deliver reliable access whenever needed. The distinction matters because assistive technology is valuable at the moment the user wants to act, not only at the moment a research protocol can produce its best result.
3. Copying a sentence and composing one are different tests
In a prompted task, the intended words are known. Researchers can compare the decoded output with the displayed sentence and calculate error. In spontaneous communication, the user chooses the message, so the intended wording may not be independently available. Both settings are important, but their accuracy measures should not be blended without explanation.
Card and colleagues' 2024 speech-neuroprosthesis paper described both an instructed copy task and a self-paced conversation mode in one participant with ALS. The author manuscript shows why the distinction is practical: controlled tasks support standardized measurement, while conversation tests whether the system can be used for messages that matter to the participant.
A hypothetical system could perform well on familiar prompted sentences but struggle with unusual names, technical vocabulary, or interrupted speech. Conversely, a participant might communicate effectively despite some word errors by using context and corrections. The useful evaluation measures both the decoding problem and the communicative outcome.
Participant ratings in everyday use add essential information, but they are not identical to an independently scored transcript with known ground truth. A report should state who assessed correctness and what categories meant. It should also describe whether the user could revise the output before the rating. These details do not invalidate natural-use evidence. They prevent an apparently comparable percentage from concealing different measurements. Everyday communication is a richer task than copying, and its evaluation should acknowledge that richness rather than forcing every result into one accuracy number.
4. Word error is not the whole cost of a wrong message
A word-level error metric treats substitutions, omissions, and insertions according to a defined calculation. It is useful for comparing decoding performance. But two sentences with the same number of word errors can have very different consequences. An error in a name, a negation, or a number may change the intended message more than an error in a redundant phrase.
Consider a hypothetical communication interface that outputs I do want that when the user intended I do not want that. The difference may involve only one short word, but the meaning reverses. An evaluation concerned with user agency should ask how such errors are detected and corrected before consequential action follows.
This is especially important when output is spoken immediately. A text interface may allow review before sending; a streaming voice can expose an error as it occurs. Neither design is universally preferable. The trade-off involves conversational timing, correction, and control. The user should have a practical way to stop, revise, or disown an unintended output.
A comprehensive report can include word-level metrics while also examining task success, correction burden, and semantic consequences. For a hypothetical email task, the outcome might be whether the intended message was sent correctly and how much effort it required. For conversation, it might include whether the partner understood the intended meaning. The technical metric remains valuable, but it should not be asked to represent every dimension of communication. Restoring a channel means restoring the person's ability to express an intention, not merely producing a high proportion of correct tokens.
5. Speed includes the pauses around the words
Words per minute is an attractive measure because it is easy to compare. But the value depends on what time is included. Counting only the period of active decoding can omit calibration, message initiation, corrections, selection of alternatives, and pauses caused by the interface. A user experiences the complete interaction.
A hypothetical system might decode at forty words per minute during a prompted sentence but require additional time to confirm and transmit it. Another might decode more slowly but permit continuous conversation with fewer interruptions. The ranking depends on whether the task is copying text, composing a message, or participating in a live exchange.
Latency matters separately from throughput. A system can produce many words per minute after a delay that makes turn-taking awkward. Littlejohn and colleagues' 2025 paper investigated continuously streaming speech synthesis in a participant with severe paralysis and anarthria. The primary abstract identifies low-delay communication as a distinct design objective, not simply another name for word accuracy.
The reporting question is therefore what the timing measure captures. Does it begin when the participant intends to speak, when a cue appears, or when the decoder starts producing output? Does it end before or after correction? Different measures can be appropriate for different questions, but they should be named precisely. A meaningful speed claim connects the metric to the interaction the user wants to have. Faster decoding is important when it translates into more usable communication, not merely a smaller interval in a selected segment of a demonstration.
6. Calibration is part of the user's day
A decoder needs a relationship between recorded signals and intended actions. Establishing and maintaining that relationship can require calibration. The amount of user effort, expert assistance, and time involved is part of the technology's practical burden. A system that performs well after an elaborate calibration session is different from one that can be started reliably with little intervention.
A hypothetical user who wants to send a short message may gain little from a high-speed decoder if preparation takes longer than the message itself. For an extended work session, the same preparation may be acceptable. Calibration burden should therefore be evaluated in relation to patterns of use rather than treated as a fixed technical inconvenience outside the outcome.
There are also different kinds of adaptation. A system may use explicit prompted tasks, learn from confirmed messages during ordinary use, or adjust automatically in the background. Each approach has advantages and possible failure modes. Learning from output that was incorrectly assumed to be correct can create a feedback problem. Requiring frequent explicit confirmation can add effort. The design needs a way to balance stability, correction, and autonomy.
Reports should state what calibration was required, who performed it, and how often it failed or had to be repeated. A reduction in calibration time can be a major advance even if peak decoding accuracy changes little. It directly affects whether the device feels like an available tool or a research session that must be arranged. The strongest system-level claim includes the path from wanting to communicate to being ready to communicate.
7. Stable performance and a stable signal are not identical
Recorded neural signals can vary over time, and a decoder may compensate for some of that variation. A stable task score does not necessarily mean the underlying signals were unchanged. Conversely, changes in recorded signals do not automatically mean the system has become unusable. The analysis should distinguish sensor behavior, decoding adaptation, and functional output.
A hypothetical system might maintain communication accuracy by updating its model frequently. That can be a successful engineering solution, but the maintenance requirement belongs in the performance description. Another system might retain a fixed decoder for long periods while producing a different balance of speed and errors. Comparing them requires more than one final accuracy number.
The relevant time horizon also matters. Stability across a session, across days, and across years are different claims. A short demonstration cannot establish durability, while a long implant duration alone does not establish continuous useful operation. The report should connect elapsed time to actual use, performance assessments, and periods of unavailability.
This distinction helps interpret progress without demanding an impossible standard of unchanging biology. A reliable assistive system can adapt. What matters is whether adaptation works transparently, preserves the user's control, and avoids an unreasonable burden. The user should not have to understand every internal update, but the research record should explain enough to determine how the observed function was sustained. Durability is a property of the whole operating system, including its capacity to respond safely when the original mapping no longer works as expected.
8. Independence needs a precise definition
Independent use can mean no researcher is present, no caregiver assistance is needed, or the user can complete selected tasks after someone else sets up the equipment. These are different achievements. Each can be valuable, but a report should specify which one was demonstrated.
The publisher's June 2026 account of the home-use study noted that care partners still required training and that broader applicability and portability remained questions. The publisher summary describes the same study as the UC Davis account, not an independent replication. Together, they support a bounded interpretation: extended home use without researcher assistance was demonstrated in one participant within a support arrangement.
A hypothetical system might allow someone to communicate privately for hours once a care partner connects external components. That can represent a substantial gain in autonomy. Calling it entirely assistance-free would obscure the setup dependency; dismissing it because setup assistance exists would ignore the independence achieved during use.
The useful evaluation breaks the workflow into stages. Who starts the system? Who resolves common faults? Who charges or connects equipment? Can the user stop output without help? Can the person choose when to use it? These questions reveal where support remains necessary and where agency has increased. Independence is not an all-or-nothing label. It is a set of capabilities that should be described in terms meaningful to the participant's daily life.
9. The comparison is with the person's available alternatives
A BCI should not be judged only against ordinary speech or hand movement. For someone with severe impairment, the relevant comparison may be the assistive options they can currently use. Those alternatives can differ in speed, fatigue, reliability, privacy, setup, and compatibility with the person's abilities. A meaningful benefit claim describes that baseline.
A hypothetical participant might already use gaze-based communication effectively in some circumstances but find it difficult in others. A neural interface could add value as an alternative or complement rather than replacing every existing tool. Another participant might have a different pattern of residual movement and different priorities. The same technical performance can therefore yield different practical benefit.
The comparison should also be fair. An unfamiliar alternative should not be tested under poor conditions simply to make the BCI look better, and the BCI should not be evaluated only after intensive expert optimization while the comparator receives little support. Training and adaptation periods may be necessary for both. The study should explain how those differences were handled.
Patient preference is important evidence, but it should be interpreted alongside task performance and burden. A participant may prefer a system because it feels more natural, allows eye contact, or preserves a sense of personal voice. Those benefits may not appear in a cursor-speed metric. The aim is to evaluate the capabilities the person values while retaining a clear account of safety and reliability. Assistive technology succeeds when it improves an actual person's options, not when it wins an abstract contest detached from those options.
10. Small studies establish possibility before generality
Early implanted-BCI studies often involve very few participants. That scale can be appropriate for feasibility work, especially when the intervention is complex and invasive. A convincing result in one person can establish that a capability is possible. It cannot establish how often the same result will occur across a broader population.
A hypothetical system might work particularly well for a participant whose neural signals, clinical history, and communication goals align with the design. Another person may require different hardware placement, training, or interface choices. The first success remains meaningful, but generalization requires evidence across relevant variation rather than assuming one person is a representative average.
This is why the number of recorded hours and the number of participants should be reported separately. Thousands of hours from one person provide rich evidence about that individual's use over time. They do not provide thousands of independent examples of how different people will respond. Longitudinal depth and population breadth answer different questions.
A strong early report makes the distinction explicit and identifies the next uncertainty. Is the priority to reproduce the capability in additional participants, improve long-term reliability, reduce support burden, or evaluate a particular risk? Progress can occur along several dimensions. The public discussion should not demand population-wide proof from a feasibility study, but it should not grant population-wide certainty because the feasibility result is moving or technically impressive. The right conclusion is often that an important possibility has been demonstrated and its range remains to be established.
11. Safety is a set of outcomes over time
An implanted BCI involves risks associated with the procedure, the device, its interfaces with the body, and its continued use. A successful communication demonstration does not measure all of those risks. Safety evaluation requires systematic follow-up and clear definitions of adverse events.
The 2023 BrainGate interim safety report analyzed fourteen adults implanted between 2004 and 2021. It reported device-related adverse events, including serious events, and described the study's specific primary safety endpoint. The primary paper provides encouraging evidence for continued research while remaining a feasibility study of a particular system and population. It should not be summarized as proof that brain implants are risk-free or that every BCI design has the same profile.
The endpoint definition matters. A study can meet a prespecified safety endpoint while still recording other important adverse events. Readers need to know what counted, the observation period, and the denominator. Implant-days, array-days, and participant counts convey different information and should not be treated as interchangeable measures of independent risk experience.
A hypothetical report of no severe events in a small cohort also has uncertainty. Rare complications may not appear simply because the study has limited exposure. That does not negate the observed safety experience; it limits how precisely risk can be estimated. The appropriate language describes what was observed and what remains unknown. Safety evidence should accumulate alongside functional evidence, rather than being inferred from the fact that a participant can use the system successfully.
12. Clinical endpoints should capture a meaningful gain
A laboratory score may show technical control without establishing a clinically meaningful improvement. For a device intended to support communication or movement, the endpoint should connect to a capability that matters in the intended population. The threshold for meaningful benefit should not be chosen solely because it makes the study easiest to declare successful.
FDA's May 2021 guidance on implanted BCIs discusses nonclinical testing and clinical considerations, including meaningful effectiveness endpoints and longer-term follow-up. The guidance is a framework for evidence development, not approval of any particular device. Its relevance is that the clinical question extends beyond whether a decoder can produce a measurable output.
A hypothetical cursor system might improve target acquisition in a standardized task. The next question is whether that gain supports useful computer access: opening applications, navigating pages, composing messages, correcting mistakes, and completing tasks without excessive fatigue. The standardized task can remain part of the evaluation, but its relationship to daily function should be demonstrated rather than assumed.
The outcome can also be multidimensional. A slower system might still improve privacy or reduce dependence on another person. A faster system might impose substantial setup or correction burden. A study should not compress these trade-offs into one favorable score without showing the components. Meaningful benefit is not merely a statistically detectable difference. It is an improvement whose scale and consequences make sense for the user and the intended role of the device.
13. The language model must not become the author
Speech BCIs may use computational language information to help convert uncertain signals into plausible words. Such assistance can improve output, but it raises a central question about agency: how much of the final message is supported by the user's signals, and how much is supplied by the system's expectations? A fluent sentence is not necessarily the intended sentence.
Consider a hypothetical decoder choosing between two phrases. A language model may favor the common phrase even when the user intended an unusual name or a surprising statement. That behavior can reduce average error while creating particular kinds of mistake. The system should allow correction and should be evaluated on messages that are not merely predictable from common language patterns.
The issue becomes more consequential if a generative assistant expands or rewrites the output. The user may welcome such assistance, but it should be a deliberate, controllable layer rather than silently attributed to neural decoding. The distinction between decoded text and machine-authored elaboration should remain visible wherever it affects consent or meaning.
A useful evaluation includes unexpected content, personal vocabulary, and the ability to reject suggestions. It asks whether the user can say something the model did not anticipate. Restoring communication requires preserving the person's authorship, including their choice to be unusual, brief, emotional, or grammatically imperfect. The goal is not to produce the most statistically likely sentence. It is to help the user express the message they intend, with a practical way to correct the system when its expectations are wrong.
14. A stop command is as important as a start command
An assistive interface needs reliable control over when output occurs. False activation can be inconvenient in a cursor task and deeply consequential in communication. The user should be able to initiate, pause, cancel, and confirm output in ways appropriate to the application. These controls are part of the core function, not optional privacy decoration.
Stanford's August 14, 2025 account of research on inner speech described proof-of-principle decoding in four participants and approaches intended to reduce unintended output, including distinguishing attempted from inner speech and a deliberate activation mechanism. The research-team interview also emphasized limits: the work did not establish accurate decoding of unconstrained private thought. It should not be described as universal mind reading.
The design question is broader than that particular study. A hypothetical system may decode intended speech accurately but occasionally activate while the user is resting or thinking silently. An evaluation should measure such events under realistic conditions, not only while the participant is explicitly performing the target task. The absence of an instruction to speak is itself an important operating state.
A reliable off state supports trust. So does an accessible way to correct or retract a message. The relevant standard is not whether a safeguard sounds reassuring in principle, but whether it works for the intended user with acceptable burden and failure behavior. Control over silence is part of control over speech.
15. Neural data and communication records need different protections
A BCI system can generate several kinds of sensitive information: raw recordings, processed features, decoder parameters, intended messages, correction histories, and usage logs. These are not all equivalent. A privacy plan should identify what is collected, what is necessary for operation, what is retained for research, and who can access each category.
A hypothetical home-use system might need recent data to improve calibration while a research project seeks longer-term recordings for scientific analysis. Those purposes can be compatible, but they should be distinguished. Consent to use an assistive device should not be treated as an unexplained blanket agreement to every future analysis of every stored signal.
The risk also depends on the capabilities of the measurement and analysis, not on a vague claim that all thoughts are recorded. Overstating what current systems can infer can frighten people without improving protection. Understating the sensitivity of longitudinal data can create a different problem. The practical approach is to assess specific data flows and plausible uses, including changes as analytical methods develop.
Users should receive understandable information about local and remote processing, access controls, retention, and options for research participation. The exact legal obligations vary and are not evaluated here. The scientific and ethical principle is narrower: data collected to restore a person's agency should not quietly undermine that agency through opaque secondary use. Privacy should be evaluated as part of system design, with boundaries tied to actual data and actual capabilities rather than promotional metaphors about reading the mind.
16. Home use is an engineering environment of its own
A laboratory can control equipment placement, technical support, and the tasks performed. A home introduces different constraints: space, movement, interruptions, caregivers, ordinary electrical equipment, and the need to use other technologies. A system that works in a home research session has begun addressing those constraints, but the degree of independence and portability still needs specification.
A hypothetical device may function well beside a dedicated equipment rack but be difficult to move between rooms. Another may be portable but require more frequent charging or have different signal reliability. These trade-offs affect which activities become possible. Portability is not merely a convenience when the intended benefit includes participation beyond a fixed workstation.
Integration with ordinary software matters too. Cursor control is useful only if the person can operate the applications they need. Changes in operating systems, accessibility settings, or communication platforms can alter usability even when the neural decoder is unchanged. The assistive system sits inside a larger technical ecosystem.
A thorough evaluation therefore includes realistic tasks and documents external dependencies. What happens after a software update? Can the user recover from a lost connection? Is an alternative communication method available during downtime? These are not reasons to dismiss early prototypes for being imperfect. They identify the engineering work between a successful experiment and a dependable service. The home is where many hidden dependencies become visible, and reporting them is a form of progress rather than an admission that the core scientific achievement did not matter.
17. Maintenance is part of durability
A durable BCI requires more than a sensor that remains implanted. External components can fail, software can change, and technical expertise can become unavailable. The system's long-term value depends on whether it can be maintained and supported through those changes. A study should distinguish biological durability from operational continuity.
A hypothetical participant may rely on a decoder that performs well but requires a specialized computer no longer supported by the research team. The neural capability has not necessarily disappeared, yet the practical communication channel may become unavailable. Conversely, replacing external components may preserve function without changing the implanted hardware. Those are different kinds of maintenance problem.
The support plan should identify responsibilities for troubleshooting, repairs, software updates, and clinical follow-up. It should also consider what happens when a participant changes location or care arrangements. A system built around one highly skilled team may be effective during a trial while requiring additional infrastructure for wider use.
Maintenance evidence can be measured. Reports can describe downtime, causes of interruption, time to restoration, and the level of expertise required. These outcomes are less dramatic than a first successful sentence, but they determine whether the device remains useful after the novelty fades. A reliable assistive technology is one that can recover from ordinary failure without leaving the user stranded. Durability should therefore be evaluated as an ongoing capacity to provide function, not only as the passage of time since implantation.
18. The end of a trial is not the end of dependence
A participant may develop meaningful reliance on an investigational device. When a study ends, questions about continued use, support, data, and possible removal remain. These questions should be considered before enrollment rather than postponed until funding or organizational circumstances force a decision.
The NIH BRAIN Initiative's May 2022 workshop on continuing trial responsibilities brought together researchers, participants, caregivers, manufacturers, insurers, regulators, and ethicists to discuss implanted neurological devices. The workshop record shows that post-trial responsibilities are a recognized part of the research problem. It does not establish one universal rule for every device or jurisdiction.
For a hypothetical participant who gains a valued communication channel, discontinuation can mean more than losing access to an experimental gadget. The device may have become part of relationships, work, and daily decision-making. That possibility does not require promising indefinite support that a team cannot deliver. It requires an honest plan describing what is known, what is uncertain, and who is responsible for decisions.
Consent should be an ongoing process as capabilities and circumstances change. Participants need information about the realistic alternatives if the study stops or the device no longer provides benefit. The ethical standard is not satisfied by a brief disclaimer that the system is experimental. The consequences of dependence deserve the same specificity as the technical performance claim. Research that aims to restore autonomy should take responsibility for how that autonomy may be affected at the end of the research relationship.
19. The participant's effort belongs in the outcome
A BCI can require concentration, repeated attempts, correction, and adaptation. A performance score may improve while the effort needed to achieve it remains high. Evaluating burden helps distinguish a capability that is possible from one the participant wants to use routinely.
A hypothetical user might complete a demanding task successfully but need substantial rest afterward. Another might prefer a slower mode because it is less tiring or permits more natural attention to a conversation partner. Neither preference can be inferred from decoding accuracy alone. The person is not simply a source of neural training data; they are the user whose experience determines practical value.
Caregiver effort also matters. A system can reduce assistance during communication while increasing setup or maintenance work. The net effect should be measured rather than assumed. A fair account includes benefits to the participant without making care partners invisible or treating all support as a failure of independence.
The study should ask about the activities people want to do, the effort they are willing to invest, and the barriers that remain. These observations can guide engineering priorities that a benchmark misses. A small improvement in setup or correction may matter more to daily use than a new peak record. Participant-centered evaluation is not a softer substitute for technical measurement. It supplies the evidence needed to decide which technical improvements actually improve life with the system.
20. The next comparison should preserve the whole system
Comparing BCIs is difficult because studies can differ in participants, sensors, tasks, vocabularies, training, and support. A headline ranking based on one metric can conceal those differences. The appropriate comparison specifies what each system was asked to do and under what conditions.
A hypothetical speech decoder evaluated on a limited prompted vocabulary should not be compared directly with an unrestricted conversation system using a different error definition. A cursor system assessed after extensive calibration should not be treated as equivalent to one evaluated immediately after startup. The results may both be impressive, but they establish different capabilities.
The same caution applies across hardware categories. Different recording approaches involve different trade-offs in invasiveness, signal characteristics, and practical use. This article does not rank them or imply that one approach is best for every person. A meaningful evaluation connects the design to an intended population and task, with safety and burden considered alongside performance.
The separation between training and evaluation is another essential comparison. In a hypothetical offline analysis, randomly dividing closely related recordings from the same session can test a narrower form of generalization than evaluating a later day. If the intended claim concerns long-term use, the evaluation should preserve the temporal challenge rather than mixing it away. The study should explain which recordings informed the decoder and which were used only to assess it. Updates during ordinary use can be legitimate, but their timing and information sources belong in the record. Otherwise, a result can appear to demonstrate future-day reliability while partly benefiting from knowledge of that day's answers. The appropriate test depends on the intended operating procedure: a system that adapts continuously should be evaluated as an adaptive system, with the permitted feedback specified, rather than being described as if it used an unchanged model throughout.
Standardized tasks can help, but they should not erase the features that matter in daily use. The ideal evidence combines comparable measurements with realistic outcomes and transparent context. That lets readers distinguish a genuine advance from a change in the test conditions. It also allows researchers to learn from complementary strengths rather than forcing every device into a single contest. The goal is a broader set of reliable options for people with different needs, not a universal winner selected by the most photogenic demonstration.
21. Progress is the ability to rely on the channel
The verified June 2026 home-use report moves the discussion toward a more consequential question: can an implanted BCI support communication and digital access as an ongoing activity rather than a scheduled demonstration? Its single-participant scope limits generalization, but the type of evidence matters. It examines function in the place where a user lives.
The next stage should make the remaining dimensions equally visible. Report accuracy and meaning, speed and delay, calibration and availability, safety and support. Distinguish researcher-free operation from assistance-free setup. Preserve the user's authorship and control over silence. Explain how data are used and what happens if the research relationship ends.
These requirements do not diminish the importance of the first cursor movement or the first decoded sentence. Those moments can establish possibilities that were previously unavailable. They become more powerful when followed by evidence that the possibility can be sustained, reproduced, and made useful without unreasonable burden.
A person using assistive technology needs more than a successful output. They need confidence that the channel will be there when they want to speak, work, ask for help, or choose not to communicate. That is the standard toward which implanted BCIs should be evaluated. The most meaningful breakthrough may be less spectacular on camera than a dramatic demonstration: an ordinary day in which the technology works reliably enough that the user can concentrate on the life they are trying to live.