Blog

Can an AI system help acquit you of drug charges? I'm not a lawyer, but just like the fictional Mike Ross taking the Bar exams without a law degree, I set out to build one anyway. Long story short, I used OpenAI models to read 97 Supreme Court cases about drug trafficking over the last 40 years:Note: The following is a screenshot of the spreadsheet with synthetic data -- here is the Google Sheets <a data-mce-href="https://docs.google.com/spreadsheets/d/16CQZhneMMO9vaFaZel-NXeS3BQv4ljjW3DdVz1v0KjQ/edit?usp=sharing" href="https://docs.google.com/spreadsheets/d/16CQZhneMMO9vaFaZel-NXeS3BQv4ljjW3DdVz1v0KjQ/edit?usp=sharing">link</a>.&nbsp;<figure id="figure-vbsufptmblwykjwco" class=""><img id="vbsufptmblwykjwco" src="https://cdn.durable.co/blocks/13iYjMNoxmq7mh69ZFeTmhSbPIOEoWv76gXsS0FF1viJ5k51nW5CSP0kCQ1rSAOb.png" alt="" title="" class="object-cover bg-gray-100 rounded-2xl md:rounded-3xl lg:rounded-4xl" style="aspect-ratio: 16/9; object-position: center center;" data-corners="default" data-aspect="16:9" data-caption="" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/13iYjMNoxmq7mh69ZFeTmhSbPIOEoWv76gXsS0FF1viJ5k51nW5CSP0kCQ1rSAOb.png" data-mce-style="aspect-ratio: 16/9; object-position: center center;"></figure>Here are some insights I learned:<h3>The Supreme Court acquitted about half of the drug trafficking cases it heard. </h3>53.61% or 52 out of 97 cases from 1984 to 2024<h3>The top five legal defenses (or as I like to call them, the loopholes) are the following:​</h3>1. Insufficiency of Evidence - used 18 times​ - accepted by the court 72.22% of the time2. Chain of Custody​ - used 46 times - acceptance rate: 71.74% 3. Denial​ - used 28 times - acceptance rate: 50.00%4. Violation of Constitutional Rights​ - used 26 times - acceptance rate: 46.15%5. Frame-up​ - used: 47 times - acceptance rate: 42.55%​<h3>However, when the Supreme Court rejects those same defenses, you are almost guaranteed of being guilty:</h3>1. Chain of Custody Rule - used 13 times in 'guilty' cases - when rejected by the court, client is guilty 100% of the time​2. Frame-up​ - used 24 times in 'guilty' cases - when rejected by the court, client is guilty 100% of the time 3. Denial​ - used 14 times in 'guilty' cases - when rejected by the court, client is guilty 85.71% of the time 4. Violation of Constitutional Rights​ - used 12 times in 'guilty' cases - when rejected by the court, client is guilty 91.66% of the time 5. Insufficiency of Evidence​ - used 12 times in 'guilty' cases - when rejected by the court, client is guilty 83.33% of the time ​The bottomline is that even though two 'loopholes' have a higher than half chance of being accepted by the court, once they are rejected, you are almost guaranteed to get sent to the locker.It's therefore extremely important that if your legal team decides to use these defenses, they need to be absolutely certain their arguments are iron-clad.This necessity now leads to the exciting part -- the AI lawyer.

How to get away with drug trafficking

an abstract image of a sphere with dots and lines

AI has tremendous opportunity for social sciences research. To demonstrate this, I am currently writing a research proposal to a research institute at my University (UP-CIDS). I am proposing an empirical study into all of jurisprudence using AI for document analysis.I was inspired by a journal article written by ANU Professor Bjoern Dressel (<a href="https://doi.org/10.1017/lsi.2022.103" data-mce-href="https://doi.org/10.1017/lsi.2022.103">https://doi.org/10.1017/lsi.2022.103</a>) that found evidence that Justices would be more likely to vote the same way as those appointed by the same President, suggesting influence of external forces:<figure id="figure-0jgnzjobglwylya0l" class=""><img id="0jgnzjobglwylya0l" src="https://cdn.durable.co/blocks/37WMn7BzNUNUX4fXkymuniadKuZ6nPcCTjOD87O9HWUlLWiFmlJXuQhHJThV5XJm.png" alt="" title="" class="object-cover bg-gray-100 rounded-2xl md:rounded-3xl lg:rounded-4xl" style="aspect-ratio: 16/9; object-position: undefined undefined;" data-corners="default" data-aspect="16:9" data-caption="" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/37WMn7BzNUNUX4fXkymuniadKuZ6nPcCTjOD87O9HWUlLWiFmlJXuQhHJThV5XJm.png" data-mce-style="aspect-ratio: 16/9; object-position: undefined undefined;"></figure>Dressel et al. selected 80 of the most prominent cases wherein political forces were most likely to have used their influence. Indeed, those appointed together were most likely to have the same opinions (concur, dissent, abstain).They measured essentially two factors for each Justice: 1. Independent Variable: the President who appointed them 2. Dependent Variable: how each Justice voted for every selected caseI then realized that an AI system would be able to expand this research into all of jurisprudence -- a dataset that I already had.Furthermore, it would be possible to measure the variables from other empirical studies cited by Dressel et al:- sociobiographic backgrounds (Tate 1970; Gatmaytan and Magno, 2011); - regime influence on court performance (Tate 1993; Haynie 1994); - ideological leanings (Pellegrina, Escresa, and Groupa, 2014); - presidential influence and judicial alignment (Escresa, 2012; Garoupa, 2013; Desierto, 2015); - qualitative dynamics (Vitug, 2010 and 2012); - loyalty effects (Epstein and Posner, 2016); - political loyalties in megapolitical cases analysis (Dressel et al., 2023)<h3>I have so far extracted decision results for about 18,000 cases -- 1901 to 1966</h3>I then converted them into a spreadsheet that looks like this:<table style="border-collapse: collapse; width: 100%;" border="1" data-mce-selected="1" data-mce-style="border-collapse: collapse; width: 100%;"><tbody><tr><td style="width: 16.665%;" data-mce-style="width: 16.665%;">G.R. No.</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Barrera</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concepcion</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Reyes</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Regala</td><td style="width: 16.6748%;" data-mce-style="width: 16.6748%;">...</td></tr><tr><td style="width: 16.665%;" data-mce-style="width: 16.665%;">L-22000</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Ponente</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Dissent</td><td style="width: 16.6748%;" data-mce-style="width: 16.6748%;">...</td></tr><tr><td style="width: 16.665%;" data-mce-style="width: 16.665%;">L-21582</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.6748%;" data-mce-style="width: 16.6748%;">...</td></tr><tr><td style="width: 16.665%;" data-mce-style="width: 16.665%;">L-21018</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Concur</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Ponente</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Dissent</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">Dissent</td><td style="width: 16.6748%;" data-mce-style="width: 16.6748%;">...</td></tr><tr><td style="width: 16.665%;" data-mce-style="width: 16.665%;">...</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">...</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">...</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">...</td><td style="width: 16.665%;" data-mce-style="width: 16.665%;">...</td><td style="width: 16.6748%;" data-mce-style="width: 16.6748%;"> </td></tr></tbody></table>I will keep on going over the next few days.The next steps would be to measure the variables defined in the previous study for every judge and every case.<h3>The long term goal is to use machine learning to compare the significance of each proposed variables.</h3>Using machine learning in the social sciences is a relatively nascent field, and there is only one paper I am aware of that uses machine learning to detect judicial bias, but not even using the same variables as in this study.I will post future updates here.

The Grand Algorithm: a survey of 80,000 Supreme Court cases

&nbsp; In the Philippines, to allow for "free and fair" bidding, all government contracts are required by law to be posted for at least 7 days on this website:<img id="ewo47b3efclwykzcu7" src="https://cdn.durable.co/blocks/14gNCRgGiMGUdQ7scUEDTfe4VUYo4ekVs2aHRR1p5G1k7M8S1MfBaRzglWj2nK8o.png" alt="" title="" class="object-cover bg-gray-100 rounded-2xl md:rounded-3xl lg:rounded-4xl" data-corners="default" data-aspect="3:2" data-caption="" data-positionx="undefined" data-positiony="undefined" data-fullwidth="true" data-mce-src="https://cdn.durable.co/blocks/14gNCRgGiMGUdQ7scUEDTfe4VUYo4ekVs2aHRR1p5G1k7M8S1MfBaRzglWj2nK8o.png" data-mce-style="aspect-ratio: 3/2; object-position: undefined undefined;" style="border-color: rgba(229,231,235,var(--tw-border-opacity)); --tw-border-spacing-x: 0; --tw-border-spacing-y: 0; --tw-translate-x: 0; --tw-translate-y: 0; --tw-rotate: 0; --tw-skew-x: 0; --tw-skew-y: 0; --tw-scale-x: 1; --tw-scale-y: 1; --tw-pan-x: ; --tw-pan-y: ; --tw-pinch-zoom: ; --tw-scroll-snap-strictness: proximity; --tw-gradient-from-position: ; --tw-gradient-via-position: ; --tw-gradient-to-position: ; --tw-ordinal: ; --tw-slashed-zero: ; --tw-numeric-figure: ; --tw-numeric-spacing: ; --tw-numeric-fraction: ; --tw-ring-inset: var(--tw-empty,/*!*/ /*!*/); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(62,172,255,.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; --tw-shadow: 0 0 #0000; --tw-shadow-colored: 0 0 #0000; --tw-blur: var(--tw-empty,/*!*/ /*!*/); --tw-brightness: var(--tw-empty,/*!*/ /*!*/); --tw-contrast: var(--tw-empty,/*!*/ /*!*/); --tw-grayscale: var(--tw-empty,/*!*/ /*!*/); --tw-hue-rotate: var(--tw-empty,/*!*/ /*!*/); --tw-invert: var(--tw-empty,/*!*/ /*!*/); --tw-saturate: var(--tw-empty,/*!*/ /*!*/); --tw-sepia: var(--tw-empty,/*!*/ /*!*/); --tw-drop-shadow: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-blur: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-brightness: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-contrast: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-grayscale: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-hue-rotate: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-invert: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-opacity: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-saturate: var(--tw-empty,/*!*/ /*!*/); --tw-backdrop-sepia: var(--tw-empty,/*!*/ /*!*/); --tw-transform: translateX(var(--tw-translate-x)) translateY(var(--tw-translate-y)) rotate(var(--tw-rotate)) skewX(var(--tw-skew-x)) skewY(var(--tw-skew-y)) scaleX(var(--tw-scale-x)) scaleY(var(--tw-scale-y)); --tw-border-opacity: 1; --tw-filter: var(--tw-blur) var(--tw-brightness) var(--tw-contrast) var(--tw-grayscale) var(--tw-hue-rotate) var(--tw-invert) var(--tw-saturate) var(--tw-sepia) var(--tw-drop-shadow); --tw-backdrop-filter: var(--tw-backdrop-blur) var(--tw-backdrop-brightness) var(--tw-backdrop-contrast) var(--tw-backdrop-grayscale) var(--tw-backdrop-hue-rotate) var(--tw-backdrop-invert) var(--tw-backdrop-opacity) var(--tw-backdrop-saturate) var(--tw-backdrop-sepia); --tw-bg-opacity: 1; width: 960px; margin: 0px; padding-top: 0px; padding-bottom: 0px; caret-color: rgba(0, 0, 0, 0); aspect-ratio: 3 / 2;">The process of finding a contract that matches your product is extremely difficult. To see the price of each contract, you have to open each of those links. To create a shortlist, you would need all day.<h3>I made a system that is able to create that shortlist almost instantly.</h3>Here is the <a href="https://docs.google.com/spreadsheets/d/1yFUvZAaF9sbYZFfHjKRaFzqluOy0Lbo_LrdGHvvIRRE/edit?usp=sharing" data-mce-href="https://docs.google.com/spreadsheets/d/1yFUvZAaF9sbYZFfHjKRaFzqluOy0Lbo_LrdGHvvIRRE/edit?usp=sharing">link </a>to my fully searchable database for contracts in Construction and Information Technology. Here's a preview:<figure id="figure-uu13heqbklwylf5pk" class=""><img id="uu13heqbklwylf5pk" src="https://cdn.durable.co/blocks/6racJwj9q4IoXRrqWG4lTytrWLWySAp3aUKGTUyRNDrA52rjukAN09YVBovcfiBm.png" alt="" title="" class="object-cover bg-gray-100 rounded-2xl md:rounded-3xl lg:rounded-4xl" style="aspect-ratio: 16/9; object-position: undefined undefined;" data-corners="default" data-aspect="16:9" data-caption="" data-positionx="undefined" data-positiony="undefined" data-fullwidth="false" data-mce-src="https://cdn.durable.co/blocks/6racJwj9q4IoXRrqWG4lTytrWLWySAp3aUKGTUyRNDrA52rjukAN09YVBovcfiBm.png" data-mce-style="aspect-ratio: 16/9; object-position: undefined undefined;"></figure>Here are some interesting insights:<h3>IT contracts (P4 billion) are already half of those involving construction (P8 billion)</h3>That is, for contracts currently posted. Each have a bidding window of about 20 days on average.<h3>Annually, IT contracts for web apps could total P2.4 billion</h3>There are currently 24 available contracts for web apps that total a little over P200 million. The contract window is 23 days on average; assuming that this is true every month, there could be P2.4 billion for web apps alone annually.<h3>Is your system enough to win over government contracts?</h3>It isn't yet. But I have also created my own database with a list of the usual awardees of the biggest acquiring entities -- some prefer a select few, while others appear to have more competitive awarding.If you were a company looking to begin government contracting, it is those more competitive acquiring entities that you should hit.

BigManilaFolder

Blog

How to get away with drug trafficking

The Grand Algorithm: a survey of 80,000 Supreme Court cases

Unveiling the shadowy world of government contracts